Understanding Non-equilibrium

Thermodynamics
G. Lebon • D. Jou • J. Casas-Vázquez
Understanding Non-equilibrium
Thermodynamics
Foundations, Applications, Frontiers
123
Prof. Dr. David Jou
Universitat Autònoma de Barcelona
Dept. Fisica - Edifici Cc
Grup Fisica Estadistica
Bellaterra 08193
Catalonia, Spain
David.Jou@uab.es
Prof. Dr. José Casas-Vázquez
Universitat Autònoma de Barcelona
Dept. Fisica - Edifici Cc
Grup Fisica Estadistica
Bellaterra 08193
Catalonia, Spain
Jose.Casas@uab.es
Prof. Dr. Georgy Lebon
Université de Liège
Dept. d’Astrophysique
G´ eophysique et Oc´ eanographie
B5 Sart Tilman
Liege 4000
Belgium
g.lebon@ulg.ac.be
ISBN: 978-3-540-74251-7 e-ISBN: 978-3-540-74252-4
Library of Congress Control Number: 2007935452
c 2008 Springer-Verlag Berlin Heidelberg
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,
reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication
or parts thereof is permitted only under the provisions of the German Copyright Law of September 9,
1965, in its current version, and permission for use must always be obtained from Springer. Violations are
liable to prosecution under the German Copyright Law.
The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply,
even in the absence of a specific statement, that such names are exempt from the relevant protective laws
and regulations and therefore free for general use.
Cover design: WMX Design GmbH, Heidelberg
Printed on acid-free paper
9 8 7 6 5 4 3 2 1
springer.com
Cover image: Thermal radiation leaving the Earth, seen by the EOS-Terra satellite
(NASA). Image from www.visibleearth.nasa.gov. Owner: NASA.
Preface
Our time is characterized by an explosion of information and by an acceler-
ation of knowledge. A book cannot compete with the huge amount of data
available on the Web. However, to assimilate all this information, it is nec-
essary to structure our knowledge in a useful conceptual framework. The
purpose of the present work is to provide such a structure for students and
researchers interested by the current state of the art of non-equilibrium ther-
modynamics. The main features of the book are a concise and critical presen-
tation of the basic ideas, illustrated by a series of examples, selected not only
for their pedagogical value but also for the perspectives offered by recent
technological advances. This book is aimed at students and researchers in
physics, chemistry, engineering, material sciences, and biology.
We have been guided by two apparently antagonistic objectives: general-
ity and simplicity. To make the book accessible to a large audience of non-
specialists, we have decided about a simplified but rigorous presentation.
Emphasis is put on the underlying physical background without sacrificing
mathematical rigour, the several formalisms being illustrated by a list of ex-
amples and problems. All over this work, we have been guided by the formula:
“Get the more from the less”, with the purpose to make a maximum of people
aware of a maximum of knowledge from a minimum of basic tools.
Besides being an introductory text, our objective is to present an overview,
as general as possible, of the more recent developments in non-equilibrium
thermodynamics, especially beyond the local equilibrium description. This
is partially a terra incognita, an unknown land, because basic concepts as
temperature, entropy, and the validity of the second law become problematic
beyond the local equilibrium hypothesis. The answers provided up to now
must be considered as partial and provisional, but are nevertheless worth to
be examined.
Chapters 1 and 2 are introductory chapters in which the main concepts
underlying equilibrium thermodynamics and classical non-equilibrium ther-
modynamics are stated. The basic notions are discussed with special emphasis
on these needed later in this book.
V
VI Preface
Several applications of classical non-equilibrium thermodynamics are pre-
sented in Chaps. 3 and 4. These illustrations have not been chosen arbitrarily,
but keeping in mind the perspectives opened by recent technological advance-
ments. For instance, advances in material sciences have led to promising
possibilities for thermoelectric devices; localized intense laser heating used to
make easier the separation of molecules has contributed to a revival of inter-
est in thermodiffusion; chemical reactions are of special interest in biology,
in relation with their coupling with active transport across membranes and
recent developments of molecular motors.
The purpose of Chaps. 5 and 6 is to discuss two particular aspects of
classical non-equilibrium thermodynamics which have been the subject of
active research during the last decades. Chapter 5 is devoted to finite-time
thermodynamics whose main concern is the competition between maximum
efficiency and maximum power and its impact on economy and ecology. This
classical subject is treated here in an updated form, taking into account the
last technological possibilities and challenges, as well as some social con-
cerns. Chapter 6 deals with instabilities and pattern formation; organized
structures occur in closed and open systems as a consequence of fluctuations
growing far from equilibrium under the action of external forces. Patterns are
observed in a multitude of our daily life experiences, like in hydrodynamics,
biology, chemistry, electricity, material sciences, or geology. After introducing
the mathematical theory of stability, several examples of ordered structures
are analysed with a special attention to the celebrated B´enard cells.
Chapters 1–6 may provide a self-consistent basis for a graduate introduc-
tory course in non-equilibrium thermodynamics.
In the remainder of the book, we go beyond the framework of the classical
description and spend some time to address and compare the most recent
developments in non-equilibrium thermodynamics. Chapters 7–11 will be of
interest for students and researchers, who feel attracted by new scientific
projects wherein they may be involved. This second part of the book may
provide the basis for an advanced graduate or even postgraduate course on
the several trends in contemporary thermodynamics.
The coexistence of several schools in non-equilibrium thermodynamics is
a reality; it is not a surprise in view of the complexity of most macroscopic
systems and the fact that some basic notions as temperature and entropy are
not univocally defined outside equilibrium. To appreciate this form of multi-
culturalism in a positive sense, it is obviously necessary to know what are the
foundations of these theories and to which extent they are related. A superfi-
cial inspection reveals that some viewpoints are overlapping but none of them
is rigorously equivalent to the other. A detailed and complete understanding
of the relationship among the diverse schools turns out to be not an easy
task. The first difficulty stems from the fact that each approach is associated
with a certain insight, we may even say an intuition or feeling that is some-
times rather difficult to apprehend. Also some unavoidable differences in the
terminology and the notation do not facilitate the communication. Another
Preface VII
factor that contributes to the difficulty to reaching a mutual comprehension
is that the schools are not frozen in time: they evolve as a consequence of
internal dynamics and by contact with others. Our goal is to contribute to
a better understanding among the different schools by discussing their main
concepts, results, advantages, and limitations. Comparison of different view-
points may be helpful for a deeper comprehension and a possible synthesis of
the many faces of the theory. Such a comparative study is not found in other
textbooks.
One problem was the selection of the main representative ones among the
wealth of thermodynamic formalisms. Here we have focused our attention
on five of them: extended thermodynamics (Chap. 7), theories with internal
variables (Chap. 8), rational thermodynamics (Chap. 9), Hamiltonian formu-
lation (Chap. 10), and mesoscopic approaches (Chap.11). In each of them, we
have tried to save the particular spirit of each theory.
It is clear that our choice is subjective: we have nevertheless been guided
not only by the pedagogical aspect and/or the impact and universality of the
different formalisms, but also by the fact that we had to restrict ourselves.
Moreover, it is our belief that a good comprehension of these different ver-
sions allows for a better and more understandable comprehension of theories
whose opportunity was not offered to be discussed here. The common points
shared by the theories presented in Chaps. 7–11 are not only to get rid of the
local equilibrium hypothesis, which is the pillar of the classical theory, but
also to propose new phenomenological approaches involving non-linearities,
memory and non-local effects, with the purpose to account for the techno-
logical requirements of faster processes and more miniaturized devices.
It could be surprising that the book is completely devoted to macroscopic
and mesoscopic aspects and that microscopic theories have been widely omit-
ted. The reasons are that many excellent treatises have been written on mi-
croscopic theories and that we decided to keep the volume of the book to a
reasonable ratio. Although statistical mechanics appears to be more fashion-
able than thermodynamics in the eyes of some people and the developments
of microscopic methods are challenging, we hope to convince the reader that
macroscopic approaches, like thermodynamics, deserve a careful attention
and are the seeds of the progress of knowledge. Notwithstanding, we remain
convinced that, within the perspectives of improvement and unification, it is
highly desirable to include as many microscopic results as possible into the
macroscopic framework.
Chapters 7–11 are autonomous and self-consistent, they have been struc-
tured in such a way that they can be read independently of each other and
in arbitrary order. However, it is highly recommended to browse through all
the chapters to better apprehend the essence and the complementarity of the
diverse theories.
At the end of each chapter is given a list of problems. The aim is not
only to allow the reader to check his understanding, but also to stimulate
his interest to solve concrete situations. Some of these problems have been
VIII Preface
inspired by recent papers, which are mentioned, and which may be consulted
for further investigation. More technical and advanced parts are confined in
boxes and can be omitted during a first reading.
We acknowledge many colleagues, and in particular M. Grmela (Montreal
University), P.C Dauby and Th. Desaive (Li`ege University), for the discus-
sions on these and related topics for more than 30 years. We also appreciate
our close collaborators for their help and stimulus in research and teach-
ing. Drs. Vicen¸ c M´endez and Vicente Ortega-Cejas deserve special gratitude
for their help in the technical preparation of this book. We also acknowl-
edge the finantial support of the Direcci´ on General de Investigaci´ on of the
Spanish Ministry of Education under grants BFM2003-06003 and FIS2006-
12296-C02-01, and of the Direcci´ o General de Recerca of the Generalitat of
Catalonia, under grants 2001 SGR 00186 and 2005 SGR 00087.
Li`ege-Bellaterra, March 2007
G. Lebon,
D. Jou,
J. Casas-V´azquez
Contents
1 Equilibrium Thermodynamics: A Review . . . . . . . . . . . . . . . . . 1
1.1 The Early History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Scope and Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 The Fundamental Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3.1 The Zeroth Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3.2 The First Law or Energy Balance . . . . . . . . . . . . . . . . . . 6
1.3.3 The Second Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.4 The Third Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4 Gibbs’ Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4.1 Fundamental Relations and State Equations . . . . . . . . . 15
1.4.2 Euler’s Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4.3 Gibbs–Duhem’s Relation . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4.4 Some Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4.5 The Basic Problem of Equilibrium Thermodynamics . . 18
1.5 Legendre Transformations and Thermodynamic Potentials . . . 19
1.5.1 Thermodynamic Potentials . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5.2 Thermodynamic Potentials and Extremum Principles . 21
1.6 Stability of Equilibrium States . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.6.1 Stability of Single Component Systems . . . . . . . . . . . . . . 24
1.6.2 Stability Conditions for the Other Thermodynamic
Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.6.3 Stability Criterion of Multi-Component Mixtures . . . . . 27
1.7 Equilibrium Chemical Thermodynamics . . . . . . . . . . . . . . . . . . . 29
1.7.1 General Equilibrium Conditions . . . . . . . . . . . . . . . . . . . . 30
1.7.2 Heat of Reaction and van’t Hoff Relation . . . . . . . . . . . . 31
1.7.3 Stability of Chemical Equilibrium and Le Chatelier’s
Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.8 Final Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
IX
X Contents
2 Classical Irreversible Thermodynamics . . . . . . . . . . . . . . . . . . . 37
2.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.2 Local Equilibrium Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.3 Entropy Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.4 General Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.5 Stationary States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.5.1 Minimum Entropy Production Principle . . . . . . . . . . . . . 51
2.6 Applications to Heat Conduction, Mass Transport,
and Fluid Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.6.1 Heat Conduction in a Rigid Body . . . . . . . . . . . . . . . . . . 54
2.6.2 Matter Diffusion Under Isothermal and Isobaric
Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.6.3 Hydrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.7 Limitations of the Classical Theory of Irreversible
Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3 Coupled Transport Phenomena . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.1 Electrical Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.2 Thermoelectric Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.2.1 Phenomenological Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.2.2 Efficiency of Thermoelectric Generators . . . . . . . . . . . . . 76
3.3 Thermodiffusion: Coupling of Heat and Mass Transport . . . . . 79
3.4 Diffusion Through a Membrane . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.4.1 Entropy Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.4.2 Phenomenological Relations . . . . . . . . . . . . . . . . . . . . . . . 85
3.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4 Chemical Reactions and Molecular Machines . . . . . . . . . . . . . 91
4.1 One Single Chemical Reaction . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.2 Coupled Chemical Reactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.2.1 General Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.2.2 Cyclical Chemical Reactions and Onsager’s
Reciprocal Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.3 Efficiency of Energy Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.4 Chemical Reactions and Mass Transport:
Molecular Machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.5 Autocatalytic Reactions and Diffusion: Morphogenesis . . . . . . 108
4.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5 Finite-Time Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.1 The Finite-Time Carnot Cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.1.1 Curzon–Ahlborn’s Model: Heat Losses . . . . . . . . . . . . . . 115
5.1.2 Friction Losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
5.2 Economical and Ecological Constraints . . . . . . . . . . . . . . . . . . . . 122
Contents XI
5.3 Earth’s Atmosphere as a Non-Equilibrium System
and a Heat Engine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5.3.1 Earth’s Energy Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5.3.2 Global Warming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5.3.3 Transformation of Solar Heat into Wind Motion . . . . . . 128
5.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6 Instabilities and Pattern Formation . . . . . . . . . . . . . . . . . . . . . . . 135
6.1 The Linear Theory of Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
6.2 Non-Linear Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.3 Thermal Convection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.3.1 The Rayleigh–B´enard’s Instability: A Linear Theory . . 145
6.3.2 The Rayleigh–B´enard’s Instability:
A Non-Linear Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.3.3 B´enard–Marangoni’s Surface Tension-Driven
Instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
6.4 Taylor’s Instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
6.5 Chemical Instabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
6.5.1 Temporal Organization in Spatially Homogeneous
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.5.2 Spatial Organization in Spatially Heterogeneous
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.5.3 Spatio-Temporal Patterns in Heterogeneous Systems:
Turing Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.6 Miscellaneous Examples of Pattern Formation . . . . . . . . . . . . . . 169
6.6.1 Salt Fingers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
6.6.2 Patterns in Electricity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
6.6.3 Dendritic Pattern Formation . . . . . . . . . . . . . . . . . . . . . . . 172
6.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
7 Extended Irreversible Thermodynamics. . . . . . . . . . . . . . . . . . . 179
7.1 Heat Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
7.1.1 Fourier’s vs. Cattaneo’s Law . . . . . . . . . . . . . . . . . . . . . . . 181
7.1.2 Extended Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
7.1.3 Non-Local Terms: From Collision-Dominated Regime
to Ballistic Regime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
7.1.4 Application to Steady Heat Transport
in Nano-Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
7.2 One-Component Viscous Heat Conducting Fluids . . . . . . . . . . . 196
7.3 Rheological Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
7.4 Microelectronic Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
7.5 Final Comments and Perspectives . . . . . . . . . . . . . . . . . . . . . . . . 205
7.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
XII Contents
8 Theories with Internal Variables . . . . . . . . . . . . . . . . . . . . . . . . . . 215
8.1 General Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
8.1.1 Accompanying State Axiom . . . . . . . . . . . . . . . . . . . . . . . 216
8.1.2 Entropy and Entropy Production . . . . . . . . . . . . . . . . . . . 219
8.1.3 Rate Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
8.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
8.2.1 Viscoelastic Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
8.2.2 Polymeric Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
8.2.3 Colloidal Suspensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
8.3 Final Comments and Comparison with Other Theories . . . . . . 232
8.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
9 Rational Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
9.1 General Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
9.2 The Axioms of Rational Thermodynamics . . . . . . . . . . . . . . . . . 238
9.2.1 Axiom of Admissibility and Clausius–Duhem’s
Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
9.2.2 Axiom of Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
9.2.3 Axiom of Equipresence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
9.2.4 Axiom of Local Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
9.2.5 Axiom of Material Frame-Indifference . . . . . . . . . . . . . . . 242
9.3 Application to Thermoelastic Materials . . . . . . . . . . . . . . . . . . . 243
9.4 Viscous Heat Conducting Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . 247
9.5 Comments and Critical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 249
9.5.1 The Clausius–Duhem’s Inequality . . . . . . . . . . . . . . . . . . 249
9.5.2 Axiom of Phlogiston . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
9.5.3 The Meaning of Temperature and Entropy . . . . . . . . . . 250
9.5.4 Axiom of Frame-Indifference . . . . . . . . . . . . . . . . . . . . . . . 251
9.5.5 The Entropy Flux Axiom. . . . . . . . . . . . . . . . . . . . . . . . . . 252
9.5.6 The Axiom of Equipresence . . . . . . . . . . . . . . . . . . . . . . . . 252
9.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
10 Hamiltonian Formalisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
10.1 Classical Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
10.2 Formulation of GENERIC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
10.2.1 Classical Navier–Stokes’ Hydrodynamics . . . . . . . . . . . . 266
10.2.2 Fickian Diffusion in Binary Mixtures . . . . . . . . . . . . . . . . 270
10.2.3 Non-Fickian Diffusion in Binary Mixtures . . . . . . . . . . . 273
10.3 Final Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
10.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
11 Mesoscopic Thermodynamic Descriptions . . . . . . . . . . . . . . . . . 279
11.1 Einstein’s Formula: Second Moments
of Equilibrium Fluctuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
11.2 Derivation of the Onsager–Casimir’s Reciprocal Relations . . . . 282
Contents XIII
11.3 Fluctuation–Dissipation Theorem. . . . . . . . . . . . . . . . . . . . . . . . . 285
11.4 Keizer’s Theory: Fluctuations in Non-Equilibrium Steady
States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
11.4.1 Dynamics of Fluctuations . . . . . . . . . . . . . . . . . . . . . . . . . 288
11.4.2 A Non-Equilibrium Entropy . . . . . . . . . . . . . . . . . . . . . . . 289
11.5 Mesoscopic Non-Equilibrium Thermodynamics . . . . . . . . . . . . . 292
11.5.1 Brownian Motion with Inertia. . . . . . . . . . . . . . . . . . . . . . 293
11.5.2 Other Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
11.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
Epilogue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
Further Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
Chapter 1
Equilibrium Thermodynamics: A Review
Equilibrium States, Reversible Processes,
Energy Conversion
Equilibrium or classical thermodynamics deals essentially with the study of
macroscopic properties of matter at equilibrium. A comprehensive definition
of equilibrium will be given later; here it is sufficient to characterize it as a
time-independent state, like a column of air at rest in absence of any flux of
matter, energy, charge, or momentum. By extension, equilibrium thermody-
namics has also been applied to the description of reversible processes: they
represent a special class of idealized processes considered as a continuum
sequence of equilibrium states.
Since time does not appear explicitly in the formalism, it would be more
appropriate to call it thermostatics and to reserve the name thermodynamics
to the study of processes taking place in the course of time outside equilib-
rium. However, for historical reasons, the name “thermodynamics” is widely
utilized nowadays, even when referring to equilibrium situations. We shall
here follow the attitude dictated by the majority but, to avoid any confu-
sion, we shall speak about equilibrium thermodynamics and designate beyond-
equilibrium theories under the name of non-equilibrium thermodynamics.
The reader is assumed to be already acquainted with equilibrium thermo-
dynamics but, for the sake of completeness, we briefly recall here the essential
concepts needed along this book. This chapter will run as follows. After a
short historical introduction and a brief recall of basic definitions, we present
the fundamental laws underlying equilibrium thermodynamics. We shall put
emphasis on Gibbs’ equation and its consequences. After having established
the criteria of stability of equilibrium, a last section, will be devoted to an
introduction to chemical thermodynamics.
1.1 The Early History
Equilibrium thermodynamics is the natural extension of the older science,
Mechanics. The latter, which rests on Newton’s law, is essentially concerned
with the study of motions of idealized systems as mass-particles and rigid
1
2 1 Equilibrium Thermodynamics: A Review
solids. Two important notions, heat and temperature, which are absent in
mechanics, constitute the pillars of the establishment of equilibrium ther-
modynamics as a branch of science. The need to develop a science beyond
the abstract approach of Newton’s law to cope with the reality of engi-
neer’s activities was born in the beginning of nineteenth century. The first
steps and concepts of thermodynamics were established by Fourier, Carnot,
Kelvin, Clausius, and Gibbs among others. Thermodynamics began in 1822
with Fourier’s publication of the Th´eorie analytique de la chaleur wherein
is derived the partial differential equation for the temperature distribution
in a rigid body. Two years later, in 1824, Sadi Carnot (1796–1832) put
down further the foundations of thermodynamics with his renowned mem-
oir R´eflexions sur la puissance motrice du feu et sur les machines propres ` a
d´evelopper cette puissance. Carnot perceived that steam power was a motor
of industrial revolution that would prompt economical and social life. Al-
though a cornerstone in the formulation of thermodynamics, Carnot’s work
is based on several misconceptions, as for instance the identification of heat
with a hypothetical indestructible weightless substance, the caloric, a notion
introduced by Lavoisier. Significant progresses towards a better comprehen-
sion of the subject can be attributed to a generation of outstanding scientists
as James P. Joule (1818–1889) who identified heat as a form of energy trans-
fer by showing experimentally that heat and work are mutually convertible.
This was the birth of the concept of energy and the basis of the formulation
of the first law of thermodynamics. At the same period, William Thomson
(1824–1907), who later matured into Lord Kelvin, realized that the work of
Carnot was not contradicting the ideas of Joule. One of his main contributions
remains a particular scale of absolute temperature. In his paper “On the dy-
namical theory of heat” appeared in 1851, Kelvin developed the point of view
that the mechanical action of heat could be interpreted by appealing to two
laws, later known as the first and second laws. In this respect, Rudolf Clausius
(1822–1888), a contemporary of Joule and Kelvin, accomplished substantial
advancements. Clausius was the first to introduce the words “internal energy”
and “entropy”, one of the most subtle notions of thermodynamics. Clausius
got definitively rid of the notion of caloric, reformulated Kelvin’s statement of
the second law, and tried to explain heat in terms of the behaviour of the indi-
vidual particles composing matter. It was the merit of Carnot, Joule, Kelvin,
and Clausius to thrust thermodynamics towards the level of an undisputed
scientific discipline. Another generation of scientists was needed to unify this
new formalism and to link it with other currents of science. One of them
was Ludwig Boltzmann (1844–1906) who put forward a decisive “mechanis-
tic” interpretation of heat transport; his major contribution was to link the
behaviour of the particles at the microscopic level to their consequences on
the macroscopic level. Another prominent scientist, Josiah Williard Gibbs
(1839–1903), deserves the credit to have converted thermodynamics into a
deductive science. In fact he recognized soon that thermodynamics of the
nineteenth century is a pure static science wherein time does not play any
1.2 Scope and Definitions 3
role. Among his main contributions, let us point out the theory of stability
based on the use of the properties of convex (or concave) functions, the po-
tential bearing his name, and the well-known Gibbs’ ensembles. Gibbs’ paper
“On the equilibrium of the heterogeneous substances” ranks among the most
decisive impacts in the developments of modern chemical thermodynamics.
Other leading scientists have contributed to the development of equilib-
rium thermodynamics as a well structured, universal, and undisputed science
since the pioneers laid down its first steps. Although the list is far from being
exhaustive, let us mention the names of Caratheodory, Cauchy, Clapeyron,
Duhem, Einstein, Helmholtz, Maxwell, Nernst, and Planck.
1.2 Scope and Definitions
Equilibrium thermodynamics is a section of macroscopic physics whose orig-
inal objective is to describe the transformations of energy in all its forms. It
is a generalization of mechanics by introducing three new concepts:
1. The concept of state, i.e. an ensemble of quantities, called state variables,
whose knowledge allows us to identify any property of the system under
study. It is desirable that the state variables are independent and easily ac-
cessible to experiments. For example, a motionless fluid may be described
by its mass m, volume V , and temperature T.
2. The notion of internal energy, complementing the notion of kinetic en-
ergy, which is of pure mechanical origin. Answering the question “what
is internal energy?” is a difficult task. Internal energy is not a directly
measurable quantity: there exist no “energymeters”. For the moment, let
us be rather evasive and say that it is presumed to be some function of
the measurable properties of a system like mass, volume, and temperature.
Considering a macroscopic system as agglomerate of individual particles,
the internal energy can be viewed as the mean value of the sum of the
kinetic and interacting energies of the particles. The notion of internal en-
ergy is also related to these temperature and heat, which are absent from
the vocabulary of mechanics.
3. The notion of entropy. Like internal energy, it is a characteristic of the
system but we cannot measure it directly, we will merely have a way to
measure its changes. From a microscopic point of view, the notion of en-
tropy is related to disorder: the higher the entropy, the larger the disorder
inside the system. There are also connections between entropy and infor-
mation in the sense that entropy can be considered as a measure of our
lack of information on the state of the system. The link between entropy
and information is widely exploited into the so-called information theory.
Energy and entropy are obeying two major laws: the first law stating that
the energy of the universe is a constant, and the second law stating that the
entropy of the universe never decreases.
4 1 Equilibrium Thermodynamics: A Review
At this stage, it is useful to recall some definitions. By system is understood
a portion of matter with a given mass, volume, and surface. An open system is
able to exchange matter and energy through its boundaries, a closed system
exchanges energy but not matter with the outside while an isolated system
does exchange neither energy nor matter with its surroundings. It is admitted
that the universe (the union of system and surroundings) acts as an isolated
system. In this chapter, we will deal essentially with homogeneous systems,
whose properties are independent of the position.
As mentioned earlier, the state of a system is defined by an ensemble of
quantities, called state variables, characterizing the system. Considering a
system evolving between two equilibrium states, A and B, it is important to
realize that, by definition, the state variables will not depend on the partic-
ular way taken to go from A to B. The selection of the state variables is not
a trivial task, and both theoretical and experimental observations constitute
a suitable guide. It is to a certain extent arbitrary and non-unique, depend-
ing on the level of description, either microscopic or macroscopic, and the
degree of accuracy that is required. A delicate notion is that of equilibrium
state which turns out to be a state, which is time independent and generally
spatially homogeneous. It is associated with the absence of fluxes of matter
and energy. On the contrary, a non-equilibrium state needs for its descrip-
tion time- and space-dependent state variables, because of exchanges of mass
and energy between the system and its surroundings. However, the above
definition of equilibrium is not complete; as shown in Sect. 1.3.3, equilibrium
of an isolated system is characterized by a maximum of entropy. Notice that
the concept of equilibrium is to some extent subjective; it is itself an idealiza-
tion and remains a little bit indefinite because of the presence of fluctuations
inherent to each equilibrium state. It depends also widely on the available
data and the degree of accuracy of our observations.
One distinguishes extensive and intensive state variables; extensive vari-
ables like mass, volume, and energy have values in a composite system equal
to the sum of the values in each subsystem; intensive variables as tempera-
ture or chemical potential take the same values everywhere in a system at
equilibrium. As a variable like temperature can only be rigorously defined
at equilibrium, one may expect difficulties when dealing with situations
beyond equilibrium.
Classical thermodynamics is not firmly restricted to equilibrium states
but also includes the study of some classes of processes, namely those that
may be considered as a sequence of neighbouring equilibrium states. Such
processes are called quasi-static and are obtained by modifying the state
variables very slowly and by a small amount. A quasi-static process is either
reversible or irreversible. A reversible process 1 →2 →3 may be viewed as a
continuum sequence of equilibrium states and will take place infinitesimally
slowly. When undergoing a reverse transformation 3 → 2 → 1, the state
variables take the same values as in the direct way and the exchanges of
matter and energy with the outside world are of opposite sign; needless to
1.3 The Fundamental Laws 5
say that reversible processes are pure idealizations. An irreversible process
is a non-reversible one. It takes place at finite velocity, may be mimicked
by a discrete series of equilibrium states and in a reverse transformation,
input of external energy from the outside is required to go back to its initial
state. Irreversible processes are generally associated with friction, shocks,
explosions, chemical reactions, viscous fluid flows, etc.
1.3 The Fundamental Laws
The first law, also popularly known as the law of conservation of energy, was
not formulated first but second after the second law, which was recognized
first. Paradoxically, the zeroth law was formulated the latest, by Fowler during
the 1930s and quoted for the first time in Fowler and Guggenheim’s book
published in 1939.
1.3.1 The Zeroth Law
It refers to the introduction of the idea of empirical temperature, which is one
of the most fundamental concepts of thermodynamics. When a system 1 is
put in contact with a system 2 but no net flow of energy occurs, both systems
are said to be in thermal equilibrium. As sketched in Fig. 1.1a, we take two
systems 1 and 2, characterized by appropriate parameters, separated by an
adiabatic wall, but in contact (a thermal contact) with the system 3 through
a diathermal wall, which allows for energy transfer in opposition with an
adiabatic wall. If the systems 1 and 2 are put in contact (see Fig. 1.1b), they
will change the values of their parameters in such a way that they reach a
state of thermal equilibrium, in which there is no net heat transfer between
them.
Fig. 1.1 Steps for introducing the empirical temperature concept
6 1 Equilibrium Thermodynamics: A Review
The zeroth law of thermodynamics states that if the systems 1 and 2 are
separately in thermal equilibrium with 3, then 1 and 2 are in thermal equilib-
rium with one another. The property of transitivity of thermal equilibrium
allows one to classify the space of thermodynamic states in classes of equiv-
alence, each of which constituted by states in mutual thermal equilibrium.
Every class may be assigned a label, called empirical temperature, and the
mathematical relation describing a class in terms of its state variables and
the empirical temperature is known as the thermal equation of state of the
system. For one mole of a simple fluid this equation has the general form
φ(p, V, θ) = 0 where p is the pressure, V the volume, and θ the empirical
temperature.
1.3.2 The First Law or Energy Balance
The first law introduces the notion of energy, which emerges as a unifying
concept, and the notion of heat, related to the transfer of energy. Here, we
examine the formulation of the first law for closed systems.
Consider first a system enclosed by a thermally isolated (adiabatic), imper-
meable wall, so that the sole interaction with the external world will appear
under the form of a mechanical work W, for instance by expansion of its
volume or by stirring. Referring to the famous experience of Joule, the work
can be measured by the decrease in potential energy of a slowly falling weight
and is given by W = mgh, where h is the displacement and g the acceler-
ation of gravity. During the evolution of the system between the two given
equilibrium states A and B, it is checked experimentally that the work W is
determined exclusively by the initial and the final states A and B, indepen-
dently of the transformation paths. This observation allows us to identify W
with the difference ∆U = U(B) − U(A) of a state variable U which will be
given the name of internal energy
W = ∆U. (1.1)
The above result provides a mean to measure the internal energy of a
system, whatever be its nature. Assume now that we remove the adiabatic
wall enclosing the system, which again proceeds from state A to state B.
When this is accomplished, it is observed that in general W ,= ∆U, and
calling Q the difference between these two quantities, one obtains
∆U −W = Q, (1.2)
where Q is referred to as the heat exchanged between the system and its sur-
roundings. Expression (1.2) is the first law of thermodynamics and is usually
written under the more familiar form
∆U = Q+W, (1.3a)
1.3 The Fundamental Laws 7
or, in terms of differentials,
dU = ¯ dQ+ ¯ dW, (1.3b)
where the stroke through the symbol “d” means that ¯ dQ and ¯ dW are inexact
differentials, i.e. that they depend on the path and not only on the initial and
final states. From now on, we adopt the sign convention that Q > 0, W > 0
when heat and work are supplied to the system, Q < 0, W < 0 when heat
and work are delivered by the system. Some authors use other conventions
resulting in a minus sign in front of ¯ dW.
It is important to stress that the domain of applicability of the first law
is not limited to reversible processes between equilibrium states. The first
law remains valid whatever the nature of the process, either reversible or
irreversible and the status of the states A and B, either equilibrium or non-
equilibrium. Designating by E = U +K+E
pot
the total energy of the system
(i.e. the sum of the internal U, kinetic K, and potential energy E
pot
), (1.3b)
will be cast in the more general form
dE = ¯ dQ+ ¯ dW. (1.4)
At this point, it should be observed that with respect to the law of energy
∆K = W as known in mechanics, we have introduced two new notions:
internal energy U and heat Q. The internal energy can be modified either
by heating the body or by acting mechanically, for instance by expansion or
compression, or by coupling both mechanisms. The quantity U consists of
a stored energy in the body while Q and W represent two different means
to transfer energy through its boundaries. The internal energy U is a state
function whose variation is completely determined by the knowledge of the
initial and final states of the process; in contrast, Q and W are not state
functions as they depend on the particular path followed by the process. It
would therefore be incorrect to speak about the heat or the work of a system.
The difference between heat and work is that the second is associated with
a change of the boundaries of the system or of the field acting on it, like a
membrane deformation or a piston displacement. Microscopically, mechanical
work is related to coherent correlated motions of the particles while heat
represents that part of motion, which is uncorrelated, say incoherent.
In equilibrium thermodynamics, the processes are reversible from which
follows that the energy balance equation (1.4) will take the form:
dU = ¯ dQ
rev
−p dV, (1.5)
wherein use is made of the classical result that the reversible work per-
formed by a piston that compress a gas of volume V and pressure p trapped
in a cylinder is given by ¯ dW
rev
= −p dV (see Problem 1.1). In engineer-
ing applications, it is customary to work with the enthalpy H defined by
H = U +pV . In terms of H, expression (1.5) of the first law reads as
dH = ¯ dQ
rev
+V dp. (1.6)
8 1 Equilibrium Thermodynamics: A Review
For an isolated system, one has simply
dU = 0 (1.7)
expressing that its energy remains constant.
Note that, when applied to open systems with n different constituents,
(1.5) will contain an additional contribution due to the exchange of matter
with the environment and takes the form (Prigogine 1947)
dU = d
i
Q−p dV +
n

k=1
h
k
d
e
m
k
; (1.8)
note that d
i
Q is not the total amount of heat but only that portion associ-
ated to the variations of the thermomechanical properties, T and p, and the
last term in (1.8), which is the extra contribution caused by the exchange
of matter d
e
m
k
with the surroundings, depends on the specific enthalpy
h
k
= H/m
k
of the various constituents.
1.3.3 The Second Law
The first law does not establish any preferred direction for the evolution of
the system. For instance, it does not forbid that heat could pass sponta-
neously from a body of lower temperature to a body of higher temperature,
nor the possibility to convert completely heat into work or that the huge
energy contained in oceans can be transformed in available work to propel
a boat without consuming fuel. More generally, the first law establishes the
equivalence between heat and work but is silent about the restrictions on the
transformation of one into the other. The role of the second law of thermo-
dynamics is to place such limitations and to reflect the property that natural
processes evolve spontaneously in one direction only. The first formulations of
the second law were proposed by Clausius (1850, 1851) and Kelvin (1851) and
were stated in terms of the impossibility of some processes to be performed.
Clausius’ statement of the second law is enunciated as follows: No process is
possible whose sole effect is to transfer heat from a cold body to a hot body.
Kelvin’s statement considers another facet: it is impossible to construct an
engine which can take heat from a single reservoir, and convert it entirely to
work in a cyclic process. In this book we will examine in detail, the formu-
lations of the second law out of equilibrium. Here, we shall concentrate on
some elements that are essential to a good understanding of the forthcoming
chapters. We will split the presentation of the second law in two parts. In
the first one, we are going to build-up a formal definition of a new quantity,
the entropy – so named by Clausius from the Greek words en (in) and trope
(turning) for representing “capacity of change or transformation” – which
is as fundamental and universal (for equilibrium systems) as the notion of
1.3 The Fundamental Laws 9
energy. In the second part, which constitutes truly the essence of the second
law, we shall enounce the principle of entropy increase during an irreversible
process.
1.3.3.1 The Concept of Entropy
Consider a homogeneous system of constant mass undergoing a reversible
transformation between two equilibrium states A and B. The quantity of
heat
_
B
A
¯ dQ
rev
depends on the path followed between states A and B (in
mathematical terms, it is an imperfect differential) and therefore cannot be
selected as a state variable. However, experimental observations have indi-
cated that by dividing ¯ dQ
rev
by a uniform and continuous function T(θ) of
an empirical temperature θ, one obtains an integral which is independent of
the path and may therefore be identified with a state function, called entropy
and denoted S
_
B
A
¯ dQ
rev
T(θ)
= ∆S = S
B
−S
A
. (1.9)
Since in reversible processes, quantities of heat are additive, entropy is also
additive and is thus an extensive quantity. A function like T(θ) which trans-
forms an imperfect differential into a perfect one is called an integrating
factor. The empirical temperature is that indicated by a mercury or an alco-
hol thermometer or a thermocouple and its value depends of course on the
nature of the thermometer; the same remark is true for the entropy, as it
depends on T(θ). It was the great merit of Kelvin to propose a temperature
scale for T, the absolute temperature, independently of any thermodynamic
system (see Box 1.1). In differential terms, (1.9) takes the form
dS =
¯ dQ
rev
T
. (1.10)
This is a very important result as it introduces two new concepts, absolute
temperature and entropy. The latter can be viewed as the quantity of heat
exchanged by the system during a reversible process taking place at the
equilibrium temperature T. Note that only differences in entropy can be
measured. Given two equilibrium states A and B, it is always possible to
determine their entropy difference regardless of whether the process between
A and B is reversible or irreversible. Indeed, it suffices to select or imagine
a reversible path joining these initial and final equilibrium states. The ques-
tion is how to realize a reversible heat transfer. Practically, the driving force
for heat transfer is a temperature difference and for reversible transfer, we
need only imagine that this temperature difference is infinitesimally small so
that ¯ dQ
rev
= lim
∆T→0
¯ dQ. Nevertheless, when the process takes place between
non-equilibrium states, the problem of the definition of entropy is open, and
actually not yet definitively solved.
10 1 Equilibrium Thermodynamics: A Review
Box 1.1 Absolute Temperature
Heat engines take heat from some hot reservoir, deliver heat to some cold
reservoir, and perform an amount of work, i.e. they partially transform heat
into work. Consider a Carnot’s reversible engine (see Fig. 1.2a) operating
between a single hot reservoir at the unknown empirical temperature θ
1
and
a single cold reservoir at temperature θ
2
. The Carnot cycle is accomplished
in four steps consisting in two isothermal and two adiabatic transformations
(Fig. 1.2b).
During the first isothermal process, the Carnot’s engine absorbs an
amount of heat Q
1
at temperature θ
1
. In the second step, the system un-
dergoes an adiabatic expansion decreasing the temperature from θ
1
to θ
2
.
Afterwards, the system goes through an isothermal compression at tem-
perature θ
2
(step 3) and finally (step 4), an adiabatic compression which
brings the system back to its initial state. After one cycle, the engine has
performed a quantity of work W but its total variation of entropy is zero
∆S
engine
=
[Q
1
[
T(θ
1
)

[Q
2
[
T(θ
2
)
= 0. (1.1.1)
Selecting the reference temperature as T(θ
2
) = 273.16, the triple point
temperature of water, it follows from (1.1.1)
T(θ
1
) = 273.16
[Q
1
[
[Q
2
[
. (1.1.2)
The ratio [Q
1
[ / [Q
2
[ is universal in the sense that it is independent of the
working substance. Therefore, Carnot cycles offer the opportunity to re-
duce temperature measurements to measurements of quantities of heat and
to define an absolute scale of positive temperatures, independently of the
measurement of temperature on any empirical temperature scale, which
depends on thermometric substance.
Fig. 1.2 (a) Heat engine and (b) Carnot diagram
1.3 The Fundamental Laws 11
The efficiency of a heat engine, in particular that of Carnot, is defined by
the ratio of the work produced to the heat supplied
η =
W
Q
1
(1.11)
for a cycle one has, in virtue of the first law, W = Q
1
−Q
2
, so that
η = 1 −
Q
2
Q
1
. (1.12)
Finally, making use of (1.1.2), it is found that the efficiency of a reversible
cycle is
η = 1 −
T
2
T
1
. (1.13)
As it will be seen, this is the maximum value for the efficiency of any heat en-
gine working between the selected heat reservoirs. More considerations about
the efficiency of reversible and irreversible cycles are developed in Chap. 5.
1.3.3.2 The Principle of Increase of Entropy
The second law was formulated by Clausius (1865) for isolated systems in
terms of the change of the entropy in the form
∆S ≥ 0. (1.14)
To illustrate the principle of entropy increase, imagine an arbitrary number of
subsystems, for instance three different gases A, B, and C at equilibrium, en-
closed in a common isolated container and separated each other by adiabatic
and rigid walls (Fig. 1.3). Let S
ini
be the entropy in this initial configuration.
Remove then the internal wall separating A and B which are diffusing into
each other until a new state of equilibrium characterized by an entropy S
int
,
corresponding to the intermediate configuration, which is larger than S
ini
is
reached. By eliminating finally the last internal constraint between A∪B and
C, and after the final state of equilibrium, corresponding to complete mixing,
is reached, it is noted that entropy S
fin
is still increased: S
fin
> S
int
> S
ini
.
Figure 1.3 reflects also that disorder is increased by passing from the initial
Fig. 1.3 Increase of entropy after removal of internal constraints
12 1 Equilibrium Thermodynamics: A Review
to the final configuration, which suggests the use of entropy as a measure of
disorder: larger the disorder larger the entropy (Bridgman 1941).
It is therefore concluded that entropy is increased as internal constraints
are removed and that entropy reaches a maximum in the final state of equi-
librium, i.e. the state of maximum “disorder”. In other terms, in isolated
systems, one has
∆S = S
fin
−S
in
≥ 0 (isolated system). (1.15)
Thus, entropy is continuously increasing when irreversible processes take
place until it reaches a state of maximum value, the equilibrium state, which
in mathematical terms is characterized by dS = 0, d
2
S < 0. This statement
constitutes the celebrated principle of entropy increase and is often referred
to as the Second Law of thermodynamics. It follows that a decrease in entropy
dS < 0 corresponds to an impossible process. Another consequence is that
the entropy of an isolated system remains constant when reversible processes
occur in it.
An illustration of the entropy increase principle is found in Box 1.2. When
the system is not isolated, as in the case of closed and open systems, the
entropy change in the system consists in two parts: d
e
S due to exchanges of
energy and matter with the outside, which may be positive or negative, and
d
i
S due to internal irreversible processes
dS = d
e
S + d
i
S. (1.16)
The second law asserts that the entropy production d
i
S can only be greater
than or equal to zero
d
i
S ≥ 0 (closed and open systems), (1.17)
the equality sign referring to reversible or equilibrium situations. Expres-
sion (1.17) is the statement of the second law in its more general form. In
the particular case of isolated systems, there is no exchange of energy and
matter so that d
e
S = 0 and one recovers (1.15) of the second law, namely
dS = d
i
S ≥ 0. For closed systems, for which d
e
S = ¯ dQ/T, one has
dS ≥ ¯ dQ/T (closed system). (1.18)
In the particular case of a cyclic process for which dS=0, one has ¯ dQ/T≤0,
which is usually identified as the Clausius’ inequality.
Box 1.2 Entropy Increase
Consider two different gases A and B at equilibrium, enclosed in a com-
mon isolated container and separated each other by an adiabatic and fixed
wall (Fig. 1.4). Both gases are characterized by their internal energy U and
volume V .
1.3 The Fundamental Laws 13
In the initial configuration, entropy S
(i)
is a function of the initial values
of internal energy U
(i)
A
and volume V
(i)
A
corresponding to subsystem A,
and similarly of U
(i)
B
and V
(i)
B
for subsystem B, in such a way that S
(i)
=
S
A
(U
(i)
A
, V
(i)
A
) + S
B
(U
(i)
B
, V
(i)
B
). If the adiabatic and fixed wall separating
both subsystems A and B is replaced by a diathermal and movable wall, a
new configuration is attained whose entropy S
(f)
may be expressed as S
(f)
=
S
A
(U
(f)
A
, V
(f)
A
) + S
B
(U
(f)
B
, V
(f)
B
); superscript (f) denotes the final values of
energy and volume submitted to the closure relations U
(i)
A
+ U
(i)
B
= U
(f)
A
+
U
(f)
B
= U
total
and V
(i)
A
+ V
(i)
B
= V
(f)
A
+ V
(f)
B
= V
total
reflecting conservation
of these quantities for the composite system A+B. The removal of internal
constraints that prevent the exchange of internal energy and volume leads
to the establishment of a new equilibrium state of entropy S
(f)
> S
(i)
.
The values taken by the (extensive) variables, in the absence of internal
constraints, in this case U
(f)
A
, V
(f)
A
and U
(f)
B
, V
(f)
B
, are those that maximize
the entropy over the manifold of equilibrium states (Callen 1985).
In Fig. 1.4 is represented S
(f)
/S
(i)
in terms of x ≡ U
A
/U
total
and y ≡
V
A
/V
total
using an ideal gas model; the final values of x and y are those
corresponding to the maximum of S
(f)
/S
(i)
. The arbitrary curve drawn on
the surface between the initial “i” and final “f” states stands for an idealized
process defined as a succession of equilibrium states, quite distinct from a
real physical process formed by a temporal succession of equilibrium and
non-equilibrium states.
Fig. 1.4 Illustration of the entropy increase principle in the case of two gases initially
separated by an adiabatic and fixed wall
14 1 Equilibrium Thermodynamics: A Review
1.3.4 The Third Law
The roots of this law appear in the study of thermodynamic quantities as
the absolute temperature tends to zero. In 1909, Nernst formulated his heat
theorem, later known as the third law of thermodynamics, to better under-
stand the nature of chemical equilibrium. Nernst’s formulation was that the
entropy change in any isothermal process approaches zero as the temperature
at which the process occurs approaches zero, i.e.
(∆S)
T→0
→0. (1.19)
This statement is sufficient for any thermodynamic development, but some-
times the stronger Planck’s statement (S → 0 as T → 0) is preferred. Since
the third law is more of quantum statistical essence, it is not of the same
nature as the other laws and no further reference will be made to it in this
book.
1.4 Gibbs’ Equation
Let us now gather the results obtained for the first and second laws. Consider
a reversible transformation, taking place in a closed system, for which the first
law takes the form
dU = ¯ dQ
rev
−p dV, (1.20)
and combine it with the definition of entropy ¯ dQ
rev
= T dS, resulting in
dU = T dS −p dV. (1.21)
Expression (1.21) is known as Gibbs’ equation; it is, however, not complete
when there are matter exchanges as in open systems, or variations in compo-
sition as in chemical reactions. To calculate the reversible work corresponding
to a chemical reaction involving n species, it is necessary to devise a reversible
process of mixing. This is achieved thanks to van’t Hoff’s box (Kestin 1968),
accordingly the reversible chemical work is given by
¯ dW
ch
rev
=
n

k=1
¯ µ
k
dm
k
, (1.22)
where ¯ µ
k
is defined as the chemical potential of substance k. The properties of
the chemical potential will be explicitly examined below. With this additional
term, one is led to the generalized Gibbs’ equations
dU = T dS −p dV +
n

k=1
¯ µ
k
dm
k
, (1.23a)
1.4 Gibbs’ Equation 15
or equivalently,
dS = T
−1
dU +pT
−1
dV −
n

k=1
T
−1
¯ µ
k
dm
k
. (1.23b)
As discussed in the forthcoming sections, the Gibbs’ equation plays a fun-
damental role in equilibrium thermodynamics. We should also mention that
Gibbs’ equation is one of the pillars of the Classical Theory of Irreversible
Processes, as shown in Chap. 2. Let us now examine the main consequences
of Gibbs’ equation.
1.4.1 Fundamental Relations and State Equations
It follows directly from Gibbs’ equation (1.23a) that
U = U(S, V, m
1
, m
2
, . . . , m
n
), (1.24)
or, solving with respect to S,
S = S(U, V, m
1
, m
2
, . . . , m
n
). (1.25)
Relations like (1.24) or (1.25) expressing that U or S are single-valued func-
tions of extensive state variables are called fundamental relations because
they contain all thermodynamic information about the system. When U (re-
spectively, S) is expressed as a function of the variables, we are speaking of
the “energy representation” (respectively, “entropy representation”).
Another consequence of Gibbs’ equation (1.23a) is that the intensive vari-
ables, represented by temperature, pressure and chemical potentials, can be
defined as partial derivatives of U:
T =
_
∂U
∂S
_
V,{m
k
}
(a), p = −
_
∂U
∂V
_
S,{m
k
}
(b), ¯ µ
k
=
_
∂U
∂m
k
_
V,S,{m
i=k
}
(c),
(1.26)
where ¦m
k
¦ stands for all m
k
constant. Since U is a function of S, V , m
k
,
the same remains true for T, p, and µ
k
so that
T = T(S, V, m
1
, m
2
, . . . , m
n
), (1.27a)
p = p(S, V, m
1
, m
2
, . . . , m
n
), (1.27b)
¯ µ
k
= ¯ µ
k
(S, V, m
1
, m
2
, . . . , m
n
). (1.27c)
Such relationships between intensive and extensive variables are called state
equations. Elimination of S between (1.27a) and (1.27b) leads to the ther-
mal equation of state p = p(T, V, m
1
, m
2
, . . . , m
n
); similarly by combining
16 1 Equilibrium Thermodynamics: A Review
(1.24) and (1.27a), one obtains the so-called caloric equation U = U(T, V, m
1
,
m
2
, . . . , m
n
). The knowledge of one single state equation is not sufficient to
describe the state of a system, which requires the knowledge of all the equa-
tions of state. For instance in the case of a monatomic perfect gas, pV = NRT
does not constitute the complete knowledge of the system but must be com-
plemented by U =
3
2
NRT, R being the gas constant and N the mole number.
1.4.2 Euler’s Relation
The extensive property of U implies that, from the mathematical point of
view, it is a first-order homogeneous function of the extensive variables:
U(λS, λV, λm
1
, . . . , λm
n
) = λU(S, V, m
1
, . . . , m
n
), (1.28)
where λ is an arbitrary scalar. Differentiation of the fundamental relation
(1.28) with respect to λ and setting λ = 1, leads to
_
∂U
∂S
_
V,{m
k
}
S +
_
∂U
∂V
_
S,{m
k
}
V +

k
_
∂U
∂m
k
_
V,S,{m
j=k
}
m
k
= U, (1.29)
and, after making use of (1.26), one obtains Euler’s relation
U = TS −pV +

k
¯ µ
k
m
k
. (1.30)
1.4.3 Gibbs–Duhem’s Relation
A differential equation among the intensive variables can be derived directly
from Euler’s relation. Indeed, after differentiating (1.30), it is found that
dU = T dS −p dV +
n

k=1
¯ µ
k
dm
k
+S dT −V dp +
n

k=1
m
k
d¯ µ
k
, (1.31)
which, after using Gibbs’ equation (1.23a), yields Gibbs–Duhem’s relation
S dT −V dp +
n

k=1
m
k
d¯ µ
k
= 0. (1.32)
It follows that the n + 2 intensive variables are not independent but related
through the Gibbs–Duhem’s relation. For a n-component mixture, the num-
ber of independent intensive state variables, called thermodynamic degrees
of freedom, is equal to n + 1: for instance, the n − 1 chemical potentials
plus temperature and pressure. In the case of a one-component fluid, the
thermodynamic description of the system requires the knowledge of two in-
dependent intensive quantities, generally selected as the temperature T and
the pressure p.
1.4 Gibbs’ Equation 17
1.4.4 Some Definitions
In view of further developments, it is useful to introduce the following defin-
itions of well-known experimental quantities:
• Coefficient of thermal expansion:
α =
1
V
_
∂V
∂T
_
p,{m
k
}
. (1.33)
• Isothermal compressibility:
κ
T
= −
1
V
_
∂V
∂p
_
T,{m
k
}
. (1.34)
• Heat capacity at constant volume:
c
V
=
_
∂U
∂T
_
V,{m
k
}
=
_
¯ dQ
rev
dT
_
V,{m
k
}
. (1.35)
• Heat capacity at constant pressure:
c
p
=
_
∂H
∂T
_
p,{m
k
}
=
_
¯ dQ
rev
dT
_
p,{m
k
}
. (1.36)
Other partial derivatives may be introduced but generally, they do not have a
specific practical usefulness. Relations between these partial derivatives may
be derived by equating mixed second-order partial derivatives of U and S.
Such expressions have been identified as Maxwell’s relations.
As a last remark, let us mention that the results established so far in
homogeneous systems of total mass m and volume V are still valid when
referred per unit mass and unit volume. Analogous to (1.24), the fundamental
relation per unit mass is
u = u(s, v, . . . , c
k
, . . .) (1.37)
with u = U/m, s = S/m, v = V/m, c
k
= m
k
/m, and

k
c
k
= 1. After
differentiation, (1.37) reads as
du = T ds −p dv +
n−1

k=1
(¯ µ
k
− ¯ µ
n
)dc
k
(1.38)
with T = (∂u/∂s)
v,{c
k
}
, p = −(∂u/∂v)
s,{c
k
}
, ¯ µ
k
= (∂u/∂c
k
)
s,v,{c
j=k
}
. Simi-
larly, the Euler and Gibbs–Duhem’s relations (1.30) and (1.32) take the form
u = Ts −pv +
n

k=1
¯ µ
k
c
k
, S dT −v dp +
n

k=1
c
k
d¯ µ
k
= 0. (1.39)
18 1 Equilibrium Thermodynamics: A Review
1.4.5 The Basic Problem of Equilibrium
Thermodynamics
To maintain a system in an equilibrium state, one needs the presence of
constraints; if some of them are removed, the system will move towards a new
equilibrium state. The basic problem is to determine the final equilibrium
state when the initial equilibrium state and the nature of the constraints
are specified. As illustration, we have considered in Box 1.3 the problem
of thermo-diffusion. The system consists of two gases filling two containers
separated by a rigid, impermeable and adiabatic wall: the whole system is
isolated. If we now replace the original wall by a semi-permeable, diathermal
one, there will be heat exchange coupled with a flow of matter between the
two subsystems until a new state of equilibrium is reached; the problem is
the calculation of the state parameters in the final equilibrium state.
Box 1.3 Thermodiffusion
Let us suppose that an isolated system consists of two separated containers
I and II, each of fixed volume, and separated by an impermeable, rigid
and adiabatic wall (see Fig. 1.5). Container I is filled with a gas A and
container II with a mixture of two non-reacting gases A and B. Substitute
now the original wall by a diathermal, non-deformable but semi-permeable
membrane, permeable to substance A. The latter will diffuse through the
membrane until the system comes to a new equilibrium, of which we want
to know the properties. The volumes of each container and the mass of
substance B are fixed:
V
I
= constant, V
II
= constant, m
B
II
= constant, (1.3.1)
but the energies in both containers as well as the mass of substance A are
free to change, subject to the constraints
U
I
+U
II
= constant, m
A
I
+m
A
II
= constant. (1.3.2)
In virtue of the second law, the values of U
I
, U
II
, m
A
I
, m
A
II
in the new equi-
librium state are such as to maximize the entropy, i.e. dS = 0 and, from
the additivity of the entropy in the two subsystems
Fig. 1.5 Equilibrium conditions for thermodiffusion
1.5 Legendre Transformations and Thermodynamic Potentials 19
dS = dS
I
+ dS
II
= 0. (1.3.3)
Making use of the Gibbs’ relation (1.23b) and the constraints (1.3.1) and
(1.3.2), one may write
dS =
∂S
I
∂U
I
dU
I
+
∂S
I
∂m
A
I
dm
A
I
+
∂S
II
∂U
II
dU
II
+
∂S
II
∂m
A
II
dm
A
II
=
_
1
T
I

1
T
II
_
dU
I

_
¯ µ
A
I
T
I

¯ µ
A
II
T
II
_
dm
A
I
= 0.
(1.3.4)
Since this relation must be satisfied for arbitrary variations of U
I
and m
A
I
,
one finds that the equilibrium conditions are that
T
I
= T
II
, ¯ µ
A
I
= ¯ µ
A
II
. (1.3.5)
The new equilibrium state, which corresponds to absence of flow of sub-
stance A, is thus characterized by the equality of temperatures and chemical
potentials in the two containers.
In absence of mass transfer, only heat transport will take place. During
the irreversible process between the initial and final equilibrium states, the
only admissible exchanges are those for which
dS =
_
1
T
I

1
T
II
_
¯ dQ
I
> 0, (1.3.6)
where use has been made of the first law dU
I
= ¯ dQ
I
. If T
I
> T
II
, one has
¯ dQ
I
< 0 while for T
I
< T
II
, ¯ dQ
I
> 0 meaning that heat will spontaneously
flow from the hot to the cold container. The formal restatement of this
item is the Clausius’ formulation of the second law: “no process is possible
in which the sole effect is transfer of heat from a cold to a hot body”.
Under isothermal conditions (T
I
= T
II
), the second law imposes that
dS =
1
T
I
_
¯ µ
A
II
− ¯ µ
A
I
_
dm
A
I
> 0 (1.3.7)
from which is concluded that matter flows spontaneously from regions of
high to low chemical potential.
1.5 Legendre Transformations and Thermodynamic
Potentials
Although the fundamental relations (1.24) and (1.25) that are expressed in
terms of extensive variables are among the most important, they are not
the most useful. Indeed, in practical situations, the intensive variables, like
20 1 Equilibrium Thermodynamics: A Review
Box 1.4 Legendre Transformations
The problem to be solved is the following: given a fundamental relation of
the extensive variables A
1
, A
2
, . . . , A
n
,
Y = Y (A
1
, A
2
, . . . , A
k
, A
k+1
, . . . , A
n
) (1.4.1)
find a new function for which the derivatives
P
i
=
∂Y
∂A
i
(i = 1, . . . , k ≤ n) (1.4.2)
will be considered as the independent variables instead of A
1
, . . . , A
k
. The
solution is given by
Y [P
1
, . . . , P
k
] = Y −
k

i=1
P
i
A
i
. (1.4.3)
Indeed, taking the infinitesimal variation of (1.4.3) results in
dY [P
1
, . . . , P
k
] = −
k

1
A
i
dP
i
+
n

k+1
P
i
dA
i
, (1.4.4)
which indicates clearly that Y [P
1
, . . . , P
k
] is a function of the indepen-
dent variables P
1
, . . . , P
k
, A
k+1
, . . . , A
n
. With Callen (1985), we have used
the notation Y [P
1
, . . . , P
k
] to denote the partial Legendre transformation
with respect to A
1
, . . . , A
k
. The function Y [P
1
, . . . , P
k
] is referred to as a
Legendre transformation.
temperature and pressure, are more easily measurable and controllable. In
contrast, there is no instrument to measure directly entropy and internal
energy. This observation has motivated a reformulation of the theory, in which
the central role is played by the intensive rather than the extensive quantities.
Mathematically, this is easily achieved thanks to the introduction of Legendre
transformations, whose mathematical basis is summarized in Box 1.4.
1.5.1 Thermodynamic Potentials
The application of the preceding general considerations to thermodynamics
is straightforward: the derivatives P
1
, P
2
, . . . will be identified with the inten-
sive variables T, −p, µ
k
and the several Legendre transformations are known
as the thermodynamic potentials. Starting from the fundamental relation,
U = U(S, V, m
k
), replace the entropy S by ∂U/∂S ≡ T as independent vari-
able, the corresponding Legendre transform is, according to (1.4.3),
1.5 Legendre Transformations and Thermodynamic Potentials 21
U[T] ≡ F = U −
_
∂U
∂S
_
V,{m
k
}
S = U −TS, (1.40)
which is known as Helmholtz’s free energy. Replacing the volume V by
∂U/∂V ≡ −p, one defines the enthalpy H as
U[p] ≡ H = U −
_
∂U
∂V
_
S,{m
k
}
V = U +pV. (1.41)
The Legendre transform which replaces simultaneously S by T and V by −p
is the so-called Gibbs’ free energy G given by
U[T, p] ≡ G = U −TS +pV =
n

k=1
m
k
¯ µ
k
. (1.42)
The last equality has been derived by taking account of Euler’s relation (1.30).
Note that the complete Legendre transform
U[T, p, µ
1
, . . . , µ
r
] = U −TS +pV −
n

k=1
¯ µ
k
m
k
= 0 (1.43)
is identically equal to zero in virtue of Euler’s relation and this explains why
only three thermodynamic potentials can be defined from U. The fundamen-
tal relations of F, H, and G read in differential form:
dF = −S dT −p dV +
n

k=1
¯ µ
k
dm
k
, (1.44a)
dH = T dS +V dp +
n

k=1
¯ µ
k
dm
k
, (1.44b)
dG = −S dT +V dp +
n

k=1
¯ µ
k
dm
k
. (1.44c)
Another set of Legendre transforms can be obtained by operating on the en-
tropy S = S(U, V, m
1
, . . . , m
n
), and are called the Massieu–Planck functions,
particularly useful in statistical mechanics.
1.5.2 Thermodynamic Potentials and Extremum
Principles
We have seen that the entropy of an isolated system increases until it attains
a maximum value: the equilibrium state. Since an isolated system does not
exchange heat, work, and matter with the surroundings, it will therefore be
22 1 Equilibrium Thermodynamics: A Review
characterized by constant values of energy U, volume V , and mass m. In
short, for a constant mass, the second law can be written as
dS ≥ 0 at U and V constant. (1.45)
Because of the invertible roles of entropy and energy, it is equivalent to for-
mulate the second principle in terms of U rather than S.
1.5.2.1 Minimum Energy Principle
Let us show that the second law implies that, in absence of any internal
constraint, the energy U evolves to a minimum at S and V fixed:
dU ≤ 0 at S and V constant. (1.46)
We will prove that if energy is not a minimum, entropy is not a maximum
in equilibrium. Suppose that the system is in equilibrium but that its in-
ternal energy has not the smallest value possible compatible with a given
value of the entropy. We then withdraw energy in the form of work, keeping
the entropy constant, and return this energy in the form of heat. Doing so,
the system is restored to its original energy but with an increased value of the
entropy, which is inconsistent with the principle that the equilibrium state is
that of maximum entropy.
Since in most practical situations, systems are not isolated, but closed and
then subject to constant temperature or (and) constant pressure, it is appro-
priate to reformulate the second principle by incorporating these constraints.
The evolution towards equilibrium is no longer governed by the entropy or
the energy but by the thermodynamic potentials.
1.5.2.2 Minimum Helmholtz’s Free Energy Principle
For closed systems maintained at constant temperature and volume, the
leading potential is Helmholtz’s free energy F. In virtue of the definition
of F(= U −TS), one has, at constant temperature,
dF = dU −T dS, (1.47)
and, making use of the first law and the decomposition dS = d
e
S + d
i
S,
dF =
¯
dQ−p dV −T d
e
S −T d
i
S. (1.48)
In closed systems d
e
S = ¯ dQ/T and, if V is maintained constant, the change
of F is
dF = −T d
i
S ≤ 0. (1.49)
1.5 Legendre Transformations and Thermodynamic Potentials 23
It follows that closed systems at fixed values of the temperature and the
volume, are driven towards an equilibrium state wherein the Helmholtz’s
free energy is minimum. Summarizing, at equilibrium, the only admissible
processes are those satisfying
dF ≤ 0 at T and V constant. (1.50)
1.5.2.3 Minimum Enthalpy Principle
Similarly, the enthalpy H = U +pV can also be associated with a minimum
principle. At constant pressure, one has
dH = dU +p dV = ¯ dQ, (1.51)
but for closed systems, ¯ dQ = T d
e
S = T(dS − d
i
S), whence, at fixed values
of p and S,
dH = −T d
i
S ≤ 0, (1.52)
as a direct consequence of the second law. Therefore, at fixed entropy and
pressure, the system evolves towards an equilibrium state characterized by a
minimum enthalpy, i.e.
dH ≤ 0 at S and p constant. (1.53)
1.5.2.4 Minimum Gibbs’ Free Energy Principle
Similar considerations are applicable to closed systems in which both tem-
perature and pressure are maintained constant but now the central quantity
is Gibbs’ free energy G = U −TS +pV . From the definition of G, one has at
T and p fixed,
dG = dU −T dS +p dV = ¯ dQ−T(d
e
S + d
i
S) = −T d
i
S ≤ 0, (1.54)
wherein use has been made of d
e
S = ¯ dQ/T. This result tells us that a closed
system, subject to the constraints T and p constant, evolves towards an equi-
librium state where Gibbs’ free energy is a minimum, i.e.
dG ≤ 0 at T and p constant. (1.55)
The above criterion plays a dominant role in chemistry because chemical
reactions are usually carried out under constant temperature and pressure
conditions.
24 1 Equilibrium Thermodynamics: A Review
It is left as an exercise (Problem 1.7) to show that the (maximum) work
delivered in a reversible process at constant temperature is equal to the de-
crease in the Helmholtz’s free energy:
¯ dW
rev
= −dF. (1.56)
This is the reason why engineers call frequently F the available work at con-
stant temperature. Similarly, enthalpy and Gibbs’ free energy are measures
of the maximum available work at constant p, and at constant T and p,
respectively.
As a general rule, it is interesting to point out that the Legendre trans-
formations of energy are a minimum for constant values of the transformed
intensive variables.
1.6 Stability of Equilibrium States
Even in equilibrium, the state variables do not keep rigorous fixed values
because of the presence of unavoidable microscopic fluctuations or external
perturbations, like small vibrations of the container. We have also seen that ir-
reversible processes are driving the system towards a unique equilibrium state
where the thermodynamic potentials take extremum values. In the particular
case of isolated systems, the unique equilibrium state is characterized by a
maximum value of the entropy. The fact of reaching or remaining in a state of
maximum or minimum potential makes that any equilibrium state be stable.
When internal fluctuations or external perturbations drive the system away
from equilibrium, spontaneous irreversible processes will arise that bring the
system back to equilibrium. In the following sections, we will exploit the con-
sequences of equilibrium stability successively in single and multi-component
homogeneous systems.
1.6.1 Stability of Single Component Systems
Imagine a one-component system of entropy S, energy U, and volume V in
equilibrium and enclosed in an isolated container. Suppose that a hypotheti-
cal impermeable internal wall splits the system into two subsystems I and II
such that S = S
I
+S
II
, U = U
I
+U
II
= constant, V = V
I
+V
II
= constant. Un-
der the action of a disturbance, either internal or external, the wall is slightly
displaced and in the new state of equilibrium, the energy and volume of the
two subsystems will take the values U
I
+∆U, V
I
+∆V, U
II
−∆U, V
II
−∆V ,
respectively; let ∆S be the corresponding change of entropy. But at equilib-
rium, S is a maximum so that perturbations can only decrease the entropy
1.6 Stability of Equilibrium States 25
∆S < 0, (1.57)
while, concomitantly, spontaneous irreversible processes will bring the sys-
tem back to its initial equilibrium configuration. Should ∆S > 0, then the
fluctuations would drive the system away from its original equilibrium state
with the consequence that the latter would be unstable. Let us now explore
the consequences of inequality (1.57) and perform a Taylor-series expansion
of S(U
I
, V
I
, U
II
, V
II
) around the equilibrium state. For small perturbations,
we may restrict the developments at the second order and write symbolically
∆S = S −S
eq
= dS + d
2
S + < 0. (1.58)
From the property that S is extremum in equilibrium, the first-order terms
vanish (dS = 0) and we are left with the calculation of d
2
S: it is found (as
detailed in Box 1.5) that
d
2
S = −T
2
C
V
(dT
−1
)
2

1
V Tκ
T
(dV )
2
< 0, (1.59)
where C
V
is the heat capacity at constant volume and κ
T
the isothermal
compressibility.
Box 1.5 Calculation of d
2
S for a Single Component System
Since the total energy and volume are constant dU
I
= −dU
II
= dU,
dV
I
= −dV
II
= dV , one may write
d
2
S =
1
2
_
_

2
S
I
∂U
2
I
+

2
S
II
∂U
2
II
_
eq
(dU)
2
+ 2
_

2
S
I
∂U
I
∂V
I
+

2
S
II
∂U
II
∂V
II
_
eq
dUdV
+
_

2
S
I
∂V
2
I
+

2
S
II
∂V
2
II
_
eq
(dV )
2
_
≤ 0. (1.5.1)
Recalling that the same substance occupies both compartments, S
I
and
S
II
and their derivatives will present the same functional dependence with
respect to the state variables, in addition, these derivatives are identical in
subsystems I and II because they are calculated at equilibrium. If follows
that (1.5.1) may be written as
d
2
S = S
UU
(dU)
2
+ 2S
UV
dUdV +S
V V
(dV )
2
≤ 0, (1.5.2)
wherein
S
UU
=
_

2
S
∂U
2
_
V
=
_
∂T
−1
∂U
_
V
, S
V V
=
_

2
S
∂V
2
_
U
=
_
∂(pT
−1
)
∂V
_
U
,
S
UV
=
_

2
S
∂U∂V
_
V
=
_
∂T
−1
∂V
_
U
=
_
∂(pT
−1
)
∂U
_
V
. (1.5.3)
26 1 Equilibrium Thermodynamics: A Review
To eliminate the cross-term in (1.5.2), we replace the differential dU by
dT
−1
, i.e.
dT
−1
= S
UU
dU +S
UV
dV, (1.5.4)
whence
d
2
S =
1
S
UU
(dT
−1
)
2
+
_
S
V V

S
2
UV
S
UU
_
(dV )
2
> 0. (1.5.5)
Furthermore, since
S
UU
=
_
∂T
−1
∂U
_
V
= −
1
T
2
_
∂T
∂U
_
V
= −
1
T
2
C
V
(1.5.6)
and
S
V V

S
2
UV
S
UU
=
_
∂(pT
−1
)
∂V
_
T
=
1
T
_
∂p
∂V
_
T
= −
1
V Tκ
T
, (1.5.7)
as can be easily proved (see Problem 1.9), (1.5.5) becomes
d
2
S = −T
2
C
V
(dT
−1
)
2

1
V Tκ
T
(dV )
2
≤ 0. (1.5.8)
The criterion (1.59) for d
2
S < 0 leads to the following conditions of stability
of equilibrium:
C
V
= (¯ dQ
rev
/dT)
V
> 0, κ
T
= −(1/V )(∂V/∂p)
T
> 0. (1.60)
The first criterion is generally referred to as the condition of thermal stability;
it means merely that, removing reversibly heat, at constant volume, must
decrease the temperature. The second condition, referred to as mechanical
stability, implies that any isothermal increase of pressure results in a diminu-
tion of volume, otherwise, the system would explode because of instabil-
ity. Inequalities (1.60) represent mathematical formulations of Le Chatelier’s
principle, i.e. that any deviation from equilibrium will induce a spontaneous
process whose effect is to restore the original situation. Suppose for exam-
ple that thermal fluctuations produce suddenly an increase of temperature
locally in a fluid. From the stability condition that C
V
is positive, and heat
will spontaneously flow out from this region (¯ dQ<0) to lower its tempera-
ture (dT <0). If the stability conditions are not satisfied, the homogeneous
system will evolve towards a state consisting of two or more portions, called
phases, like liquid water and its vapour. Moreover, when systems are driven
far from equilibrium, the state is no longer characterized by an extremum
principle and irreversible processes do not always maintain the system stable
(see Chap. 6).
1.6 Stability of Equilibrium States 27
1.6.2 Stability Conditions for the Other
Thermodynamic Potentials
The formulation of the stability criterion in the energy representation is
straightforward. Since equilibrium is characterized by minimum energy, the
corresponding stability criterion will be expressed as d
2
U(S, V ) ≥ 0 or, more
explicitly,
U
SS
≥ 0, U
V V
≥ 0, U
SS
U
V V
−(U
SV
)
2
≥ 0 (1.61)
showing that the energy is jointly a convex function of U and V (and also of
N in open systems).
The results are also easily generalized to the Legendre transformations
of S and U. As an example, consider the Helmholtz’s free energy F. From
dF = −S dT −p dV , it is inferred that
F
TT
= −T
−1
C
V
≤ 0, F
V V
=
1
V κ
T
≥ 0 (1.62)
from which it follows that F is a concave function of temperature and a convex
function of the volume as reflected by the inequalities (1.62). By concave
(convex) function is meant a function that lies everywhere below (above)
its family of tangent lines, be aware that some authors use the opposite
definition for the terms concave and convex. Similar conclusions are drawn
for the enthalpy, which is a convex function of entropy and a concave function
of pressure:
H
SS
≥ 0, H
pp
≤ 0. (1.63a)
Finally, the Gibbs’ free energy G is jointly a concave function of temperature
and pressure
G
TT
≤ 0, G
pp
≤ 0, G
TT
G
pp
−(G
Tp
)
2
≥ 0. (1.63b)
1.6.3 Stability Criterion of Multi-Component Mixtures
Starting from the fundamental relation of a mixture of n constituents in
the entropy representation, S = S(U, V, m
1
, m
2
, . . . , m
n
), it is detailed in
Box 1.6 that the second-order variation d
2
S, which determines the stability,
is given by
d
2
S = dT
−1
dU + d(pT
−1
)dV −
n

k=1
d(¯ µ
k
T
−1
)dm
k
≤ 0. (1.64)
At constant temperature and pressure, inequality (1.64) reduces to
n

k,l
∂¯ µ
k
∂m
l
dm
k
dm
l
≥ 0, (1.65)
28 1 Equilibrium Thermodynamics: A Review
Box 1.6 Calculation of d
2
S for Multi-Component Systems
In an N-component mixture of total energy U, total volume V , and total
mass m =

k
m
k
the second variation of S is
d
2
S =
1
2
(S
UU
)
eq
(dU)
2
+
1
2
(S
V V
)
eq
(dV )
2
+
1
2

k,l
(S
m
k
m
l
)
eq
dm
k
dm
l
+ (S
UV
)
eq
dUdV +

k
(S
Um
k
)
eq
dUdm
k
+

k
(S
V,m
k
)
eq
dV dm
k
.
(1.6.1)
Making use of the general results ∂S/∂U = 1/T, ∂S/∂V = p/T, ∂S/∂m
k
=
−µ
k
/T, the above expression can be written as
d
2
S =
1
2
_
∂T
−1
∂U
dU +
∂T
−1
∂V
dV +

k
∂T
−1
∂m
k
dm
k
_
dU
+
1
2
_
∂(pT
−1
)
∂U
dU +
∂(pT
−1
)
∂V
dV +

k
∂(pT
−1
)
∂m
k
_
dV

1
2

k
_
∂(¯ µ
k
T
−1
)
∂U
dU +
∂(¯ µ
k
T
−1
)
∂V
dV +

l
∂(¯ µ
k
T
−1
)
∂m
l
dm
l
_
dm
k
(1.6.2)
from which follows the general stability condition
d
2
S =
1
2
_
dT
−1
dU + d(pT
−1
)dV −
n

k=1
d(¯ µ
k
T
−1
)dm
k
_
≤ 0. (1.6.3)
to be satisfied whatever the values of dm
k
and dm
l
from which follows that:
∂¯ µ
k
∂m
k
≥ 0, det
¸
¸
¸
¸
∂¯ µ
k
∂m
l
¸
¸
¸
¸
≥ 0. (1.66)
The criteria (1.65) or (1.66) are referred to as the conditions of stability
with respect to diffusion. The first inequality (1.66) indicates that the stabil-
ity of equilibrium implies that any increase on mass of a given constituent
will increase its chemical potential. This provides another example of the
application of Le Chatelier’s principle. Indeed, any non-homogeneity which
manifests in a part of the system in the form of increase of mass will induce
locally an increase of chemical potential. Since the latter is larger than its
ambient value, there will be a net flow of matter from high to lower chemical
potentials that will tend to eradicate the non-homogeneity.
1.7 Equilibrium Chemical Thermodynamics 29
1.7 Equilibrium Chemical Thermodynamics
In the last part of this chapter, we shall apply the general results of equi-
librium thermodynamics to chemical thermodynamics. As an illustration,
consider the reaction of synthesis of hydrogen chloride
H
2
+ Cl
2
2HCl (1.67)
or more generally
n

k=1
ν
k
X
k
= 0, (1.68)
where the X
k
s are the symbols for the n chemical species and ν
k
the stoi-
chiometric coefficients; conventionally the latter will be counted positive when
they correspond to products and negative for reactants. In the above example,
X
1
= H
2
, X
2
= Cl
2
, X
3
= HCl and ν
1
= −1, ν
2
= −1, ν
3
= 2, n = 3. The
reaction may proceed in either direction depending on temperature, pressure,
and composition; in equilibrium, the quantity of reactants that disappear is
equal to the quantity of products that instantly appear. The change in the
mole numbers dN
k
of the various components of (1.68) is governed by
dN
H
2
−1
=
dN
Cl
2
−1
=
dN
HCl
2
≡ dξ, (1.69)
where ξ is called the degree of advancement or extent of reaction. At the
beginning of the reaction ξ = 0; its time derivative dξ/dt is related to the
velocity of reaction which vanishes at chemical equilibrium (see Chap. 4).
The advantage of the introduction of ξ is that all the changes in the mole
numbers are expressed by one single parameter, indeed from (1.69),
dN
k
= ν
k
dξ, (1.70)
and, after integration,
N
k
= N
0
k

k
ξ, (1.71)
wherein superscript 0 denotes the initial state; observe that the knowledge
of ξ completely specifies the composition of the system. When expressed in
terms of the mass of the constituents, (1.70) reads as dm
k
= ν
k
M
k
dξ with M
k
the molar mass of k; after summation on k, one obtains the mass conservation
law
n

k=1
ν
k
M
k
= 0. (1.72)
The above results are directly generalized when r chemical reactions are
taking place among the n constituents. In this case, (1.70) and (1.72) will,
respectively, be of the form
30 1 Equilibrium Thermodynamics: A Review
dN
k
=
r

j=1
ν
jk

j
(k = 1, 2, . . . , n),
n

k=1
ν
jk
M
k
= 0 (j = 1, 2, . . . , r).
(1.73)
In Sects. 1.7.1–1.7.3, we discuss further the conditions for chemical equilib-
rium and the consequences of stability of equilibrium.
1.7.1 General Equilibrium Conditions
Since chemical reactions take generally place at constant temperature and
pressure, it is convenient to analyse them in terms of Gibbs’ free energy
G = G(T, p, N
1
, N
2
, . . . , N
n
). At constant temperature and pressure, the
change in G associated with the variations dN
k
in the mole numbers is
dG =
n

k=1
µ
k
dN
k
, (1.74)
where µ
k
is the chemical potential per mole of species k. This µ
k
is closely
related to that appearing in (1.23), because the mass of the species k is
m
k
= M
k
N
k
, with M
k
the corresponding molar mass. Then, it is immediate
to see that µ
k
= ¯ µ
k
/M
k
. Since µ
k
and ¯ µ
k
are often found in the literature,
it is useful to be acquainted with both of them. After substitution of dN
k
by
its value (1.70), one has
dG =
_
n

k=1
ν
k
µ
k
_
dξ = −/dξ, (1.75)
wherein, with De Donder, we have introduced the “affinity” of the reaction
as defined by
/ = −
n

k=1
ν
k
µ
k
. (1.76)
Since G is a minimum at equilibrium (dG/dξ = 0), the condition of chemical
equilibrium is that the affinity is zero:
/
eq
= 0. (1.77)
In presence of r reactions, equilibrium implies that the affinity of each in-
dividual reaction vanishes: (/
j
)
eq
= 0(j = 1, 2, . . . , r). To better apprehend
the physical meaning of the result (1.77), let us express / in terms of phys-
ical quantities. In the case of ideal gases or diluted solutions, the chemical
potential of constituent k can be written as
µ
k
= µ
0
k
(p, T) +RT ln x
k
, (1.78)
1.7 Equilibrium Chemical Thermodynamics 31
where x
k
= N
k
/N is the mole fraction of substance k, N the total number
of moles and µ
0
k
(p, T) is the part of chemical potential depending only on p
and T. For non-ideal systems, the above form is preserved at the condition to
replace ln x
k
by ln(x
k
γ
k
) where γ
k
is called the activity coefficient and a
k
=
γ
k
x
k
the activity. By substituting (1.78) in expression (1.76) of the affinity,
we can express the equilibrium condition in terms of the mole fractions, which
are measurable quantities, and one has
/
eq

k
(−ν
k
µ
0
k
) −RT

k
ν
k
ln x
k
= 0. (1.79)
Defining the equilibrium constant K(T, p) by means of ln K(T, p) =
−(

k
ν
k
µ
0
k
)/RT, the previous relation can be cast in the simple form
/
eq
= RT ln
K(T, p)
x
ν
1
1
x
ν
2
2
x
ν
n
n
= 0, (1.80)
whence
x
ν
1
1

eq
)x
ν
2
2

eq
) x
ν
n
n

eq
) = K(T, p), (1.81)
which is called the mass action law or Guldberg and Waage law. This is the
key relation in equilibrium chemistry: it is one algebraic equation involving
a single unknown, namely, the value of ξ
eq
of degree of advancement which
gives the corresponding number of moles in equilibrium through (1.71). If
K(T, p) is known as a function of T and p for a particular reaction, all the
equilibrium mole fractions can be computed by the mass action law. Going
back to our illustrative example (1.67) and assuming that each component is
well described by the ideal gas model, the law of mass action is
x
2
HCl
x
H
2
x
Cl
2
= K(T, p). (1.82)
When a number r of reactions are implied, we will have r algebraic equations
of the form (1.81) to be solved for the unknowns (ξ
1
)
eq
, (ξ
2
)
eq
, . . . , (ξ
r
)
eq
.
1.7.2 Heat of Reaction and van’t Hoff Relation
Most of the chemical reactions supply or absorb heat, thus the heat of reaction
is an important notion in chemistry. To introduce it, let us start from the
first law written as
¯ dQ = dH −V dp, (1.83)
with the enthalpy H given by the equation of state H = H(T, p, ξ), whose
differential form is
dH =
_
∂H
∂T
_
p,ξ
dT +
_
∂H
∂p
_
T,ξ
dp +
_
∂H
∂ξ
_
T,p
dξ. (1.84)
32 1 Equilibrium Thermodynamics: A Review
Substitution of (1.84) in (1.83) yields
¯ dQ = C
p,ξ
dT +h
T,ξ
dp −r
T,p
dξ, (1.85)
where C
p
= (∂H/∂T)
p,ξ
, h
T
= (∂H/∂p)
T,ξ
− V , and r
T,p
= −(∂H/∂ξ)
T,p
designate the specific heat at constant pressure, the heat compressibility and
the heat of reaction at constant temperature and pressure, respectively. The
heat of reaction is positive if the reaction is exothermic (which corresponds to
delivered heat) and negative if the reaction is endothermic (which corresponds
to absorbed heat). In terms of variation of affinity with temperature, the heat
of reaction is given by
r
T,p
= −T
_
∂/
∂T
_
p,ξ
eq
. (1.86)
This is directly established by using the result
H = G+TS = G−T
_
∂G
∂T
_
p,ξ
. (1.87)
From the definition of the heat of reaction, one has
r
T,p
= −
_
∂G
∂ξ
_
T,p
+T
_

∂T
_
∂G
∂ξ
_
T,p
_
p,ξ
, (1.88)
with (∂G/∂ξ)
T,p
= −/ in virtue of Gibbs’ equation (1.75), substituting this
result in (1.88) and recognizing that /
eq
= 0, one obtains (1.86). By means
of (1.78) of µ
k
and (1.79) of /, it is easily found from (1.86) that
r
T,p
= −RT
2

∂T
ln K(T, p). (1.89)
This is the van’t Hoff relation, which is very important in chemical ther-
modynamics; it permits to determine the heat of reaction solely from the
measurements of the equilibrium constant K(T, p) at different temperatures.
1.7.3 Stability of Chemical Equilibrium
and Le Chatelier’s Principle
For the clarity of the presentation, let us recall the condition (1.65) for sta-
bility of diffusion which can be cast in the form:
n

k,l
µ
kl
dN
k
dN
l
> 0, µ
kl
=
∂µ
k
∂N
l
. (1.90)
1.7 Equilibrium Chemical Thermodynamics 33
Assuming that there are no simultaneous reactions and inserting (1.70), one
obtains
n

k,l
µ
kl
ν
k
ν
l
(dξ)
2
> 0. (1.91)
From the other side, it follows from the definition of the affinity that
∂/
∂ξ
= −
n

k,l
ν
k
∂N
k
∂N
l
∂N
l
∂ξ
= −
n

k,l
µ
kl
ν
k
ν
l
, (1.92)
so that the criterion of stability (1.90) can be cast in the simple form

∂/
∂ξ
(dξ)
2
> 0 or
∂/
∂ξ
< 0. (1.93)
The above result is easily generalized for r simultaneous reactions:
r

m,n
∂/
m
∂ξ
n

m

n
< 0 (m, n = 1, 2, . . . , r). (1.94)
Perhaps one of the most interesting consequences of the stability criterion
(1.93) is in the form of Le Chatelier’s moderation principle. Let us first ex-
amine the effect on chemical equilibrium of a temperature change at constant
pressure. The shift of equilibrium with temperature is measured by the quan-
tity (∂ξ/∂T)
p,A
, index / is introduced here because in chemical equilibrium,
/ is constant, in fact zero. It is a mathematical exercise to prove that (see
Problem 1.8)
_
∂ξ
∂T
_
p,A
= −
(∂//∂T)
p,ξ
(∂//∂ξ)
p,T
. (1.95)
The numerator is related to the heat of reaction by (1.86) from which follows
that:
_
∂ξ
∂T
_
p,A
=
1
T
r
T,p
(∂//∂ξ)
p,T
. (1.96)
According to the stability condition (1.93), the denominator is a negative
quantity so that the sign of (∂ξ/∂T)
p,A
is opposite to the sign of the heat
of reaction r
T,p
. Therefore, an increase of temperature at constant pressure
will shift the reaction in the direction corresponding to endothermic reaction
(r
T,p
< 0). This is in the direction in which heat is absorbed, thus opposing
the increase of temperature. Similar results are obtained by varying the pres-
sure: an increase of pressure at constant temperature will cause the reaction
to progress in the direction leading to a diminution of volume, thus weak-
ening the action of the external effect. These are particular examples of the
more general principle of Le Chatelier stating that any system in chemical
equilibrium undergoes, under the effect of external stimuli, a compensating
change which will be always in the opposite direction.
34 1 Equilibrium Thermodynamics: A Review
1.8 Final Comments
Equilibrium thermodynamics constitutes a unique and universal formalism
whose foundations are well established and corroborated by experience. It
has also been the subject of numerous applications. It should nevertheless be
kept in mind that equilibrium thermodynamics is of limited range as it deals
essentially with equilibrium situations and idealized reversible processes. It is
therefore legitimate to ask to what extent equilibrium thermodynamics can
be generalized to cover more general situations as non-homogeneous systems,
far from equilibrium states and irreversible processes. Many efforts have been
spent to meet such objectives and have resulted in the developments of various
approaches coined under the generic name of non-equilibrium thermodynam-
ics. It is our purpose in the forthcoming chapters to present, to discuss, and
to compare the most recent and relevant – at least in our opinion – of these
beyond of equilibrium theories.
There exists a multiplicity of excellent textbooks on equilibrium thermo-
dynamics and it would be unrealistic to go through the complete list. Let
us nevertheless mention the books by Callen (1985), Duhem (1911), Gibbs
(1948), Kestin (1968), Prigogine (1947), Kondepudi and Prigogine (1998) and
Zemansky (1968), which have been a source of inspiration for the present
chapter.
1.9 Problems
1.1. Mechanical work. Starting from the mechanical definition of work
¯ dW = F dx (scalar product of force and displacement), show that the
work done during the compression of a gas of volume V is ¯ dW = −p dV , and
that the same expression is valid for an expansion.
1.2. Carnot cycle. Show that the work performed by an engine during an
irreversible cycle operating between two thermal reservoirs at temperatures
T
1
and T
2
< T
1
is given by W = W
max
−T
2
∆S, where ∆S is the increase of
entropy of the Universe, and W
max
is the corresponding work performed in
a reversible Carnot cycle.
1.3. Fundamental relation. In the entropy representation, the fundamental
equation for a monatomic ideal gas is
S(U, V, N) =
N
N
0
S
0
+NRln
_
_
U
U
0
_
3/2
_
V
V
0
__
N
N
0
_
−5/2
_
,
with R the ideal gas constant, and the subscript 0 standing for an arbitrary
reference state. By using the formalism of equilibrium thermodynamics, show
that the thermal and caloric equations of state for this system are pV = NRT
and U =
3
2
NRT, respectively.
1.9 Problems 35
1.4. Maxwell’s relations. The equality of the second crossed derivatives of the
thermodynamic potentials is a useful tool to relate thermodynamic quantities.
Here, we will consider some consequences related to second derivatives of the
entropy. (a) Show that dS(T, V ) at constant N may be expressed as
dS =
1
T
_
∂U
∂T
_
V
dT +
1
T
__
∂U
∂V
_
T
+p
_
dV.
(b) By equating the second crossed derivatives of S in this expression, show
that
p +T
_
∂p
∂T
_
V
=
_
∂U
∂V
_
T
.
(c) Using this equation, show that for ideal gases, for which pV = NRT,
the internal energy does not depend on V , i.e. (∂U/∂V )
T
= 0. (d) For elec-
tromagnetic radiation, p = (1/3)(U/V ). Using this result and the relation
obtained in (b), show that for that system, one has U = aV T
4
. Remark:
This expression is closely related to the Stefan–Boltzmann law for the heat
flux radiated by a black body, namely q = σT
4
, with σ the Stefan–Boltzmann
constant, indeed it follows from the relation q = 1/4c(U/V ), with c the speed
of light, and a given by a = 4σ/c.
1.5. Maxwell’s relations. Prove that c
p
− c
v
= Tvα
2

T
by making use of
Maxwell’s relations.
1.6. Van der Waals gases. The thermal equation of state for real gases was
approximated by van der Waals in the well-known expression
_
p +a
_
N
V
_
2
_
(V −bN) = NRT,
where a and b are positive constants, fixed for each particular gas, which are,
respectively, related to the attractive and repulsive intermolecular forces and
are null for ideal gases. (a) Using the expressions derived in Problem 1.4,
show that the caloric equation of state U = U(T, V, N) has the form
U(T, V, N) = U
id
(T, N) −a
N
2
V
,
with U
id
(T, N) is the internal energy for ideal gases. (b) Calculate the change
of temperature in an adiabatic expansion against the vacuum, i.e. the vari-
ation of T in terms of V at constant internal energy. (c) Find the curve
separating the mechanically stable region from the mechanically unstable re-
gion in the plane p − V , the mechanical stability condition being given by
(1.60). (d) The maximum of such a curve, defined by the additional con-
dition (∂
2
p/∂V
2
)
T,N
= 0, is called the critical point. Show that the val-
ues of p, V , and T at the critical point are p
c
= (1/27)(a/b
2
), V
c
= 3bN,
T
c
= (8/27)(a/bR), respectively.
36 1 Equilibrium Thermodynamics: A Review
1.7. Kelvin’s statement of the second law. Verify that in a reversible trans-
formation at fixed temperature and volume, the maximum reversible work
delivered by the system is equal to the decrease of the Helmholtz’s free energy
−dF. As an aside result, show that it is not possible to obtain work from
an engine operating in a cycle and in contact with one single source of heat.
This result is known as Kelvin’s statement of the second law.
1.8. Mathematical relation. Verify the general result
_
∂y
∂x
_
f,z
= −
(∂f/∂x)
y,z
(∂f/∂y)
x,z
.
Hint: Consider f = f(X, Y, Z) = constant as an implicit function of the
variables X, Y, Z and write explicitly df = 0.
1.9. Stability coefficient. Prove that (see Box 1.5)

2
S
∂V
2

(∂
2
S/∂U∂V )
2

2
S/∂U
2
=
1
T
_
∂p
∂V
_
T
.
Hint: Construct the Massieu function (i.e. Legendre transformation of en-
tropy), namely S[T
−1
] = S − T
−1
U. From the differential of S[T
−1
], derive
(∂
2
S[T
−1
]/∂V
2
)
T
.
1.10. Stability conditions. Reformulate the stability analysis of Sect. 1.6.1 by
considering that the total entropy and volume are kept constant and by
expanding the total energy in Taylor’s series around equilibrium.
1.11. Le Chatelier’s principle. Show that an increase in pressure, at constant
temperature, causes the chemical reaction to proceed in that direction which
decreases the total volume.
1.12. Le Chatelier’s principle. The reaction of dissociation of hydrogen iodide
2HI →H
2
+I
2
is endothermic. Determine in which direction equilibrium will
be shifted when (a) the temperature is decreased at constant pressure and
(b) the pressure is decreased at constant temperature.
Chapter 2
Classical Irreversible Thermodynamics
Local Equilibrium Theory of Thermodynamics
Equilibrium thermodynamics is concerned with ideal processes taking place
at infinitely slow rate, considered as a sequence of equilibrium states. For
arbitrary processes, it may only compare the initial and final equilibrium
states but the processes themselves cannot be described. To handle more
realistic situations involving finite velocities and inhomogeneous effects, an
extension of equilibrium thermodynamics is needed.
A first insight is provided by the so-called “classical theory of irreversible
processes” also named “classical irreversible thermodynamics” (CIT). This
borrows most of the concepts and tools from equilibrium thermodynamics
but transposed at a local scale because non-equilibrium states are usually in-
homogeneous. The objective is to cope with non-equilibrium situations where
basic physical quantities like mass, temperature, pressure, etc. are not only
allowed to change from place to place, but also in the course of time. As shown
in the present and the forthcoming chapters, this theory has been very use-
ful in dealing with a wide variety of practical problems. Pioneering works
in this theory were accomplished by Onsager (1931) and Prigogine (1961);
these authors were awarded the Nobel Prize in Chemistry in 1968 and 1977,
respectively. Other important and influential contributions are also found in
the works of Meixner and Reik (1959), de Groot and Mazur (1962), Gyarmati
(1970), and many others, which have enlarged the theory to a wider number
of applications and have clarified its foundations and its limits of validity.
The principal aims of classical irreversible thermodynamics are threefold.
First to provide a thermodynamic support to the classical transport equations
of heat, mass, momentum, and electrical charge, as the Fourier’s law (1810)
relating the heat flux to the temperature gradient, Fick’s relation (1850)
between the flux of matter and the mass concentration gradient, Ohm’s equa-
tion (1855) between electrical current and potential, and Newton–Stokes’
law (1687, 1851) relating viscous pressure to velocity gradient in fluids. A
second objective is to propose a systematic description of the coupling be-
tween thermal, mechanical, chemical, and electromagnetic effects, as the Soret
(1879) and Dufour (1872) effects, coupling heat, and mass transport, and the
37
38 2 Classical Irreversible Thermodynamics
Seebeck (1821), Peltier (1836) and Thomson effects, coupling thermal trans-
port, and electric current. A third objective is the study of stationary non-
equilibrium dissipative states, whose properties do not depend on time, but
which are characterized by a non-homogeneous distribution of the variables
and non-vanishing values of the fluxes.
The present chapter is divided in two parts: in the first one, we recall
briefly the general statements underlying the classical theory of irreversible
processes. The second part is devoted to the presentation of a few simple
illustrations, as heat conduction in rigid bodies, matter transport, and hydro-
dynamics. Chemical reactions and coupled transport phenomena, like ther-
moelectricity, thermodiffusion, and diffusion through a membrane are dealt
with separately in Chaps. 3 and 4.
2.1 Basic Concepts
The relevance of transport equations, which play a central role in non-
equilibrium thermodynamics – comparable, in some way, to equations of state
in equilibrium thermodynamics – justifies some preliminary considerations.
Transport equations describe the amount of heat, mass, electrical charge, or
other quantities which are transferred per unit time between different sys-
tems and different regions of a system as a response to a non-homogeneity in
temperature T, molar concentration c, electric potential ϕ
e
. Historically, the
first incursions into this subject are allotted to Fourier, Fick, and Ohm, who
proposed the nowadays well-known laws:
q = −λ∇T (Fourier’s law), (2.1)
J = −D∇c (Fick’s law), (2.2)
I = σ
e
∇ϕ
e
(Ohm’s law). (2.3)
Here q is the heat flux (amount of internal energy per unit time and unit
area transported by conduction), J is the diffusion flux (amount of matter,
expressed in moles, transported per unit time and unit area), and I is the
flux of electric current (electric charge transported per unit time and area).
The coefficients λ, D, and σ
e
are the thermal conductivity, diffusion coeffi-
cient, and electric conductivity, respectively. The knowledge of these various
transport coefficients in terms of temperature, pressure, and mass concentra-
tion has important consequences in material sciences and more generally on
our everyday life. For instance, a low value of thermal conductivity is needed
for a better isolation of buildings; in contrast, large values are preferred to
avoid excessive heating of computers; the diffusion coefficient is a fundamen-
tal parameter in biology and in pollution dispersal problems, while electrical
conductivity has a deep influence on the development and management of
electrical plants, networks, and microelectronic devices. The value of some
2.2 Local Equilibrium Hypothesis 39
transport coefficients, as for instance electrical conductivity, may present
huge discontinuities as, for example, in superconductors, where conductiv-
ity diverges, thus implying a vanishing electrical resistivity, or in insulators,
where conductivity vanishes.
The physical content of the above transport laws is rather intuitive: heat
will flow from regions with higher temperature to regions at lower temper-
ature and the larger is the temperature gradient the larger is the heat flow.
Analogously, matter diffuses from regions with higher mass concentration to
regions with lower concentration, and electric positive charges move from
regions with higher electrical potential to regions with lower potential.
The evolution of a system in the course of time and space requires the
knowledge of the balance between the ingoing and the outgoing fluxes. If the
outgoing flow is larger than the incoming one, the amount of internal energy,
number of moles, or electric charge in the system will increase, and it will
decrease if the situation is reversed. When ingoing and outgoing fluxes are
equal, the properties of the system will not change in the course of time, and
the system is in a non-equilibrium steady state. In equilibrium, all fluxes van-
ish. It should be noticed that the above considerations apply only for so-called
conserved quantities, which means absence of production or consumption in-
side the system. When source terms are present, as for instance in chemical
reactions, one should add new contributions expressing the amount of moles,
which is produced or destroyed.
Expressions (2.1)–(2.3) of the classical transport laws were originally pro-
posed either from theoretical considerations or on experimental grounds. As
stated before, non-equilibrium thermodynamics aims to propose a general
scheme for the derivation of the transport laws by ensuring that they are
compatible with the laws of thermodynamics (for instance, thermal conduc-
tivity must be positive because, otherwise, heat would spontaneously flow
from lower to higher temperature, in conflict with the second law). Indeed,
when λ, D, and σ
e
are scalar quantities it is relatively evident that they
must be positive; however, when they are tensors (as in anisotropic systems
or in the presence of magnetic fields) the thermodynamic restrictions on the
values of their components are not obvious and to obtain them a careful
and systematic analysis is required. It opens also the way to the study of
coupled situations where there are simultaneously non-homogeneities in tem-
perature, concentration, and electric potential, instead of considering these
effects separately as in (2.1)–(2.3).
2.2 Local Equilibrium Hypothesis
The most important hypothesis underlying CIT is undoubtedly the local
equilibrium hypothesis. According to it, the local and instantaneous relations
between thermodynamic quantities in a system out of equilibrium are the same
40 2 Classical Irreversible Thermodynamics
as for a uniform system in equilibrium. To be more explicit, consider a system
split mentally in a series of cells, which are sufficiently large for microscopic
fluctuations to be negligible but sufficiently small so that equilibrium is real-
ized to a good approximation in each individual cell. The size of such cells has
been a subject of debate, on which a good analysis can be found in Kreuzer
(1981) and Hafskjold and Kjelstrup (1995).
The local equilibrium hypothesis states that at a given instant of time,
equilibrium is achieved in each individual cell or, using the vocabulary of
continuum physics, at each material point. It should, however, be realized
that the state of equilibrium is different from one cell to the other so that,
for example, exchanges of mass and energy are allowed between neighbouring
cells. Moreover, in each individual cell the equilibrium state is not frozen
but changes in the course of time. A better description of this situation is
achieved in terms of two timescales: the first, τ
m
, denotes the equilibration
time inside one cell and it is of the order of the time interval between two
successive collisions between particles, i.e. 10
−12
s, at normal pressure and
temperature. The second characteristic time τ
M
is a macroscopic one whose
order of magnitude is related to the duration of an experiment, say about 1 s.
The ratio between both reference times is called the Deborah number De =
τ
m

M
. For De ¸1, the local equilibrium hypothesis is fully justified because
the relevant variables evolve on a large timescale τ
M
and do practically not
change over the time τ
m
, but the hypothesis is not appropriate to describe
situations characterized by De ¸1. Large values of De are typical of systems
with long relaxation times, like polymers, for which τ
m
may be of the order
of 100 s, or of high-frequency or very fast phenomena, such as ultrasound
propagation, shock waves, nuclear collisions, for which τ
M
is very short, say
between 10
−5
and 10
−10
s.
A first consequence of the local equilibrium assumption is that all the
variables defined in equilibrium as entropy, temperature, chemical potential,
etc. are univocally defined outside equilibrium, but they are allowed to vary
with time and space. Another consequence is that the local state variables
are related by the same state equations as in equilibrium. This means, in
particular, that the Gibbs’ relation between entropy and the state variables
remains locally valid for each value of the time t and the position vector r.
For example, in the case of a n-component fluid of total mass m, the local
Gibbs’ equation will be written as
ds = T
−1
du +pT
−1
dv −T
−1
n

k=1
µ
k
dc
k
, (2.4)
where s is the specific entropy (per unit mass), u is the specific internal
energy, related to the specific total energy e by u = e −
1
2
v v, with v the
velocity field of the centre of mass of the cell, T is the absolute temperature,
p is the hydrostatic pressure, v is the specific volume related to the mass
density ρ by v = 1/ρ, c
k
= m
k
/m, the mass fraction of substance k, and µ
k
its chemical potential.
2.3 Entropy Balance 41
A third consequence follows from the property that, locally, the system
is stable. Therefore, in analogy with equilibrium situations, such quantities
as heat capacity, isothermal compressibility or the Lam´e coefficients in the
theory of elasticity are positive definite.
More generally, Gibbs’ equation will take the form
ds(r, t) =

i
Γ
i
(r, t) da
i
(r, t), (2.5)
where a
i
(r, t) is an extensive state variable, like u, v, c
k
, while Γ
i
(r, t) is the
corresponding conjugate intensive state variable, for instance T, p or µ
k
.
The above relation is assumed to remain valid when expressed in terms of
the material (or Lagrangian) time derivative d/dt = ∂/∂t + v ∇, i.e. by
following a small cell moving with velocity v:
ds(r, t)
dt
=

i
Γ
i
(r, t)
da
i
(r, t)
dt
. (2.6)
From the kinetic theory point of view, the local equilibrium hypothesis is jus-
tified only for conditions where the Maxwellian distribution is approximately
maintained. Otherwise, it should be generalized, as indicated in the second
part of this book.
2.3 Entropy Balance
An important question is whether a precise definition can be attached to
the notion of entropy when the system is driven far from equilibrium. In
equilibrium thermodynamics, entropy is a well-defined function of state only
in equilibrium states or during reversible processes. However, thanks to the
local equilibrium hypothesis, entropy remains a valuable state function even
in non-equilibrium situations. The problem of the definition of entropy and
corollary of intensive variables as temperature will be raised as soon as the
local equilibrium hypothesis is given up.
By material body (or system) is meant a continuum medium of total mass
m and volume V bounded by a surface Σ. Consider an arbitrary body, outside
equilibrium, whose total entropy at time t is S. The rate of variation of this
extensive quantity may be written as the sum of the rate of exchange with
the exterior d
e
S/dt and the rate of internal production, d
i
S/dt:
dS
dt
=
d
e
S
dt
+
d
i
S
dt
. (2.7)
As in Chap. 1, we adopt the convention that any quantity (like mass,
energy, entropy) is counted positive if supplied to the system and negative if
transferred from it to the surroundings. The quantity Td
i
S/dt is sometimes
42 2 Classical Irreversible Thermodynamics
called the uncompensated heat or the rate of dissipation. For further purpose,
it is convenient to introduce the notion of entropy flux J
s
, i.e. the entropy
crossing the boundary surface per unit area and unit time, and the rate of
entropy production σ
s
, i.e. the entropy produced per unit volume and unit
time inside the system. In terms of these quantities, d
e
S/dt and d
i
S/dt may
be written as
d
e
S
dt
= −
_
Σ
J
s
n dΣ, (2.8)
d
i
S
dt
=
_
V
σ
s
dV , (2.9)
in which n is the unit normal pointing outwards the volume of the body.
Once entropy is defined, it is necessary to formulate the second law, i.e.
to specify which kinds of behaviours are admissible in terms of the entropy
behaviour. The classical formulation of the second law due to Clausius states
that, in isolated systems, the possible processes are those in which the entropy
of the final equilibrium state is higher or equal (but not lower) than the
entropy of the initial equilibrium state. In the classical theory of irreversible
processes, one introduces an even stronger restriction by requiring that the
entropy of an isolated system must increase everywhere and at any time, i.e.
dS/dt ≥ 0. In non-isolated systems, the second law will take the more general
form
d
i
S/dt > 0 (for irreversible processes), (2.10a)
d
i
S/dt = 0 (for reversible processes or at equilibrium). (2.10b)
It is important to realize that inequality (2.10a) does nor prevent that open or
closed systems driven out of equilibrium may be characterized by dS/dt < 0;
this occurs for processes for which d
e
S/dt < 0 and larger in absolute value
than d
i
S/dt. Several examples are discussed in Chap. 6.
After introducing the specific entropy s in such a way that S =
_
ρs dV
and using the definitions (2.8) and (2.9), the entropy balance (2.7) reads as
d
dt
_
ρs dV = −
_
Σ
J
s
n dΣ +
_
V
σ
s
dV . (2.11)
In virtue of the Gauss and Reynolds theorems, the above equation takes the
form
_
ρ
ds
dt
dV = −
_
V
∇ J
s
dV +
_
V
σ
s
dV , (2.12)
where ∇ ≡ (∂/∂x, ∂/∂y, ∂/∂z) designates the nabla operator whose compo-
nents are the partial space derivatives in Cartesian coordinates. Assuming
that (2.12) is valid for any volume V and that the integrands are continuous
functions of position, one can write the following local balance relation
ρ
ds
dt
= −∇ J
s

s
, (2.13)
2.3 Entropy Balance 43
with, in virtue of the second law as stated by ((2.10a) and (2.10b)),
σ
s
≥ 0, (2.14)
where the equality sign refers to reversible processes. This quantity is impor-
tant in engineering because the product Tσ
s
is a measure of the degradation
or dissipation of energy in engines, and its minimization may be useful to
enhance their efficiency.
Finally, it is worth to mention that the second law introduces an asym-
metry in time for irreversible processes, which is often known as an arrow of
time, an interesting physical and even philosophical concept that is briefly
presented in Box 2.1.
Box 2.1 Irreversibility and the Arrow of Time
An interesting aspect of the principle of entropy increase is the special light
it sheds on the concept of time, a view that departs radically from classical
mechanics. The equations of Newtonian mechanics are deterministic and
reversible with respect to time reversal. Time is considered as an external
parameter, which describes the chronology of a succession of events; it is the
time given by our watch. Reversible processes do not distinguish between
the future and the past. In contrast, the principle of increase of entropy
makes possible the distinction between future and past as it implies an arrow
of time, which imposes that irreversible processes proceed spontaneously
within a given direction in time. It establishes a fundamental anisotropy
in Nature and provides a criterion allowing to decide whether a process is
going forwards or backwards.
In relation with the notions of time, reversibility and irreversibility
was raised the following problem. At the microscopic level, the motion of
the individual particles composing the macroscopic systems is described
by Newton’s equation and is therefore of reversible nature, whereas at the
macroscopic level, the systems behave irreversibly. The antagonism between
these two behaviours has been a major source of debate since the micro-
scopic interpretation of entropy by Boltzmann, in the years 1870. A wide-
spread view is that the behaviour of the systems is intrinsically reversible
but that the large number of particles makes that actually, it evolves towards
an irreversible dynamics due to our inability to follow each individual parti-
cle. In other words, irreversibility is an illusion raised up by our ignorance.
Following this attitude, irreversibility should not be related to the system
itself but to the observer.
Irreversibility is also associated with loss of information, a view rein-
forced by the information theory. Accordingly, the entropy is interpreted as
a lack of information about the microscopic state of the system, and loss
of information means higher entropy. When entropy production is large,
much information is lost per unit time, whereas in reversible processes no
information is lost. The strong interrelations between information theory
44 2 Classical Irreversible Thermodynamics
and thermodynamics have been underlined in many books and papers (e.g.
Shannon 1948; Jaynes 1963; Keizer 1987).
In the 1970s, the study of deterministic chaos has opened new per-
spectives. Accordingly, though the equations of mechanics are reversible
and deterministic, the sensibility of their solutions with respect to slight
changes of the initial conditions or by varying some external parameters as
temperature gradients (see Chap. 6) makes that the predictability is lost.
Interesting conceptual discussions on this subject have been put forward
by Prigogine (e.g. Prigogine and Stengers 1979; Prigogine 1996), who ar-
gues that irreversibility is present at all levels of description, microscopic
and macroscopic, and merges out, as a source of order, from the instabil-
ities present in the system. The breaking of time symmetry is introduced
by constructing an appropriate operator playing the role of entropy and
monotonically increasing in time. The formalism is coined under the name
of subdynamics. Prigogine’s line of thought is not unanimously accepted
and target of keen discussions (see, e.g. Thom 1972; Bricmont 1995; Van
Kampen 2000).
The above considerations show that the problem of evolution of re-
versibility towards irreversibility has not received a definitive answer yet
(Lebowitz 1999). A last remark will concern ageing, which is a sensible
problem to everybody and assuredly the most visible irreversible face of
life. Experiences of our daily life tell us that youth will not return. There
are many theories of biological ageing but these topics are outside the scope
of the present book. It seems rather evident that ageing is related to the
entropy production, or the amount of irreversibility, inside the metabolic
processes, and that small entropy production should produce slow ageing.
More considerations about the convergence of non-equilibrium thermody-
namics and biology may be found in Chap. 4.
2.4 General Theory
Our main objective is to present a macroscopic description of irreversible
processes and our major task is to determine the evolution equations of the
relevant local variables, in accordance with the fundamental laws of thermo-
dynamics. To give a general idea of the contents and assumptions underlying
classical irreversible thermodynamics, we consider a system outside equilib-
rium and characterized by a finite set of variables to be specified later on.
To make clear the general structure, we will proceed in several steps. The
same steps, as introduced in this section, will be systematically repeated in
the future.
2.4 General Theory 45
Step 1. Space of state variables
The choice of state variables is determined by three arguments:
1. It depends on the system under consideration and the degree of accuracy
that one wishes to achieve.
2. The variables must be “dynamically admissible”, which means that their
timescale is sufficiently different from the timescales corresponding to the
more microscopic levels of description.
3. Theoretical predictions should be in agreement with experimental obser-
vations, and in addition, the results should be reproducible. This does not
mean that the state variables are directly measurable but it is required to
express them in terms of measured quantities.
According to the local equilibrium hypothesis, the space of the state vari-
ables is the ensemble a(r, t) = [a
1
(r, t), a
2
(r, t), . . . , a
N
(r, t)] of the exten-
sive thermodynamic variables appearing in Gibbs’ equation, plus the velocity
field v(r, t), i.e. the union of a mechanical variable (the barycentric velocity
of each subcell) and thermodynamic variables (i.e. each subcell is considered
as a thermodynamic system described by the usual set of thermodynamic
quantities as in equilibrium). In the case of an n-component fluid, the ther-
modynamic variables are simply the specific volume v(r, t), the mass fraction
c
k
(r, t)(k = 1, 2, . . ., n) of the n constituents and the internal energy u(r, t).
Step 2. Evolution equations
They take the form of balance equations written in whole generality as
ρ
da
dt
= −∇ J
a

a
, (2.15)
where J
a
is the flux term expressing the exchange with the surroundings and
σ
a
is the corresponding source term; when this term is zero, a is conserved.
To be explicit, and to identify the flux and source terms, let us write down the
balance equations of total mass, mass concentrations, energy and momentum
for a n-constituent mixture (de Groot and Mazur 1962):
ρ
dv
dt
= ∇ v, (2.16)
ρ
dc
k
dt
= −∇ J
k

k
, (2.17)
ρ
du
dt
= −∇ q −P
T
: ∇v +ρr, (2.18)
ρ
dv
dt
= −∇ P+ρF, (2.19)
where the superscript T means transposition. In (2.16), the flux term is simply
the velocity and there is no source because of conservation of mass. In (2.17)–
(2.19), the fluxes are the diffusion flux J
k
of substance k, the heat flux q
and the pressure tensor P, respectively; the corresponding sources are the
46 2 Classical Irreversible Thermodynamics
rate of production σ
k
of substance k by chemical reactions, the mechanical
work −P
T
: (∇v) plus some possible energy source r per unit mass and the
external body force F per unit mass. The main difference between q and
r is that q measures the transfer of heat across the bounding surface while
r is the energy supply distributed within the volume. We refer to P as the
pressure tensor because pure hydrostatic pressure corresponds to P = pI,
where I is the identity tensor with components δ
ij

ij
= 1 for i = j and
δ
ij
= 0 for i ,= j); the colon in (2.18) stands for the double scalar product
A : B = A
ij
B
ji
. Einstein’s summation convention on repeated indices, when
they refer to Cartesian coordinates, will be used throughout the book.
The set (2.16)–(2.19) describes the evolution of the system, i.e. how the
basic properties change with time at every point in space. Note that it con-
tains more unknown quantities than equations, because the fluxes q and P
are generally not known and have to be expressed in terms of the basic vari-
ables to close the system. In that respect the second law is very useful as
shown by the next step.
Step 3. Entropy production and second law of thermodynamics
It is the purpose to examine the consequences resulting from the entropy
inequality σ
s
> 0. Substitution of the evolution equations (2.15) of the basic
variables in the Gibbs’ relation (2.6) leads to an explicit equation for the
entropy evolution which, after comparison with the general expression (2.13)
of the entropy balance, allows us to identify J
s
and σ
s
. The latter is found
to consist of a sum of products of so-called thermodynamic fluxes J
α
and
thermodynamic forces X
α
:
σ
s
=

α
J
α
X
α
. (2.20)
Note that the thermodynamic forces are not forces in the mechanical sense,
but they are quantities generally related to the gradients of the intensive
variables whereas the fluxes J
α
can be identified with the fluxes of energy,
mass, momentum, etc. In (2.20), the fluxes and forces may be scalars, vectors,
or tensors, and the product J
α
X
α
stands indifferently for the usual product
between two scalars, the scalar product between two vectors and the double
scalar product between two tensors. Explicit expressions for J
α
and X
α
will be
given later on while treating applications. It is important to point out that at
equilibrium (or for reversible processes), the thermodynamic fluxes and forces
vanish identically so that entropy production is zero in such situations, as it
should.
Step 4. Linear flux–force relations
Experience indicates that the thermodynamic fluxes and forces are not inde-
pendent but that there exists a relationship between them. Moreover, it has
been observed that for a large class of irreversible processes, the fluxes are
linear functions of the forces, to a good approximation. This is furthermore
the simplest way to ensure that the rate of entropy production is a positive
quantity. Within the hypothesis of linearity, one has
2.4 General Theory 47
J
α
=

β
L
αβ
X
β
. (2.21)
These flux–force relations are named phenomenological relations, constitu-
tive or transport equations, they express the relation between causes (the
forces) and effects (the fluxes) and the specific properties of the materials
under study. At equilibrium, both members of (2.21) vanish identically. The
phenomenological coefficients L
αβ
are generally depending on the intensive
variables T, p, and c
k
; the coefficient L
αα
connects a flow J
α
to its conjugate
force X
α
while L
αβ
describes the coupling between two irreversible processes
α and β. In thermoelectricity for instance, L
αα
is related to the electrical
resistance and L
αβ
to the coupling between the electric current and the heat
flow. Simple examples of linear phenomenological relations are the Fourier’s,
Fick’s, and Ohm’s equations (2.1)–(2.3).
Step 5. Restrictions due to material symmetry: Curie’s law
According to relation (2.21), one should in principle be allowed to couple any
flux to any force. However, material symmetry reduces the number of cou-
plings between fluxes and forces. This property is known in CIT as Curie’s
law. It reflects the property that macroscopic causes cannot have more ele-
ments of symmetry than the effects they produce. This restriction plays an
important role, especially in isotropic systems, for which the properties at
equilibrium are the same in all directions.
Truly, as shown in Box 2.2, it is evident that Curie’s law is nothing
more than an application of the representation theorems of isotropic tensors
(Spencer and Rivlin 1959; Truesdell and Noll 1965). To illustrate this state-
ment, consider an isotropic body subject to a hypothetical irreversible process
described by the fluxes j (a scalar), J (a vector), T (a symmetric tensor of
order 2). In addition, the fluxes j and J are assumed to depend on the ther-
modynamic forces x (a vector) and X(a tensor of order 2) but Tdepends only
on X.
As a consequence of isotropy, and under the hypothesis of linear flux–force
relations (see Box 2.2), one obtains the following phenomenological relations:
j = l tr X, (2.22a)
J = A
1
x, (2.22b)
T = B
1
(tr X)I +B
2
X, (2.22c)
where the phenomenological scalar coefficients l, A
1
, B
1
, and B
2
are inde-
pendent of x and X, and tr X denotes the trace of tensor X. The results
((2.22a), (2.22b), and (2.22c)) exhibit the property that, in isotropic systems
and within the linear regime, it is forbidden to couple fluxes and forces of
different tensorial character. For instance a chemical affinity (a scalar) can-
not give raise to a heat flux (a vector), similarly a temperature gradient
(a vector) is unable to induce a mechanical stress (a tensor of order 2).
48 2 Classical Irreversible Thermodynamics
Box 2.2 Curie’s Law
The existence of spatial symmetries in a material system contributes to sim-
plify the scheme of the phenomenological relations. Because of invariance of
the phenomenological equations under special orthogonal transformations,
some couplings between fluxes and forces are not authorized in isotropic
systems. In the examples treated in this chapter, no fluxes and forces of ten-
sorial order higher than the second will occur. We shall therefore consider
an isotropic material characterized by the three following phenomenological
relations
j = j(x, X), J = J(x, X), T = T(X), (2.2.1)
which contain fluxes and forces of tensorial order 0 (the scalar j), order
1 (the vectors J and x) and order 2 (the symmetric tensors T and X).
Isotropy imposes that the above relations transform as follows under an
orthogonal transformation Q:
j(Q x; Q X Q
T
) = j(x, X), (2.2.2)
J(Q x; Q X Q
T
) = Q J(x, X), (2.2.3)
T(Q X Q
T
) = Q T(X) Q
T
. (2.2.4)
According to the theorems of representation of isotropic tensors (e.g.
Spencer and Rivlin 1959), the functions j, J, and T are isotropic if and
only if
j(x, X) = j[I
X
, II
X
, III
X
, x x, x (X x), x (X X x)], (2.2.5)
J(x, X) = (A
1
I +A
2
X+A
3
X X) x, (2.2.6)
T(X) = B
1
I +B
2
X+B
3
X X, (2.2.7)
the flux j and the coefficients A
i
(i = 1, 2, 3) are isotropic scalar functions
of x and X and the B
i
are isotropic scalars of X alone; I
X
, II
X
, and III
X
are the principal invariants of the tensor X, namely
I
X
= tr X, II
X
=
1
2
[I
2
X
−tr(X X)], III
X
= det X. (2.2.8)
By restricting the analysis to linear laws, (2.2.5)–(2.2.7) will take the simple
form
j = l(tr X), J = A
1
x, T = B
1
(tr X)I +B
2
X, (2.2.9)
wherein l, A
1
, B
1
, and B
2
are now scalars independent of x and X. Re-
lations (2.2.9) exhibit the property that in linear constitutive equations,
spatial symmetry allows exclusively the coupling between fluxes and forces
of the same tensorial order, as concluded from Curie’s law.
2.4 General Theory 49
Step 6. Restrictions on the sign of phenomenological coefficients
A direct restriction on the sign of the phenomenological coefficients arises as a
consequence of the second law. Substitution of the linear flux–force relations
(2.21) into (2.20) of the rate of entropy production yields the quadratic form
σ
s
=

αβ
L
αβ
X
α
X
β
≥ 0. (2.23)
According to standard results in algebra, the necessary and sufficient condi-
tions for σ
s
≥ 0 are that the determinant [L
αβ
+L
βα
[ and all its principal
minors are non-negative. It follows that
L
αα
≥ 0, (2.24)
while the cross-coefficients L
αβ
must satisfy
L
αα
L
ββ

1
4
(L
αβ
+L
βα
)
2
. (2.25)
In virtue of inequality (2.24), all the transport coefficients like the heat con-
ductivity, the diffusion coefficient, and the electrical resistance are positive,
meaning that heat flows from high to low temperature, electrical current
from high to low electric potential, and neutral solutes from higher to lower
concentrations.
Step 7. Restrictions on L
αβ
due to time reversal: Onsager–Casimir’s
reciprocal relations
It was established by Onsager (1931) that, besides the restrictions on the sign,
the phenomenological coefficients verify symmetry properties. The latter were
presented by Onsager as a consequence of “microscopic reversibility”, which
is the invariance of the microscopic equations of motion with respect to time
reversal t →−t. Accordingly, by reversing the time, the particles retrace their
former paths or, otherwise stated, there is a symmetry property between
the past and the future. Invoking the principle of microscopic reversibility
and using the theory of fluctuations, Onsager was able to demonstrate the
symmetry property
L
αβ
= L
βα
. (2.26)
In Chaps. 4 and 11, we will present detailed derivations of (2.26).
It should, however, be stressed that the above result holds true only for
fluctuations a
α
(t) = A
α
(t) −A
eq
α
of extensive state variables A
α
with respect
to their equilibrium values, which are even functions of time a
α
(t) = a
α
(−t).
In the case of odd parity of one of the variables, α or β, for which a
α
(t) =
−a
α
(−t), the coefficients L
αβ
are skew-symmetric instead of symmetric, as
shown by Casimir (1945), i.e.
L
αβ
= −L
βα
. (2.27)
In a reference frame rotating with angular velocity ω and in presence of an
external magnetic field H, Onsager–Casimir’s reciprocal relations take the
form
50 2 Classical Irreversible Thermodynamics
L
αβ
(ω, H) = ±L
βα
(−ω, −H), (2.28)
as it will be shown on dynamical bases in Chap. 11. The validity of the
Onsager–Casimir’s reciprocal relations is not limited to phenomenological
transport coefficients that are scalar quantities as discussed earlier. Consider
for example an irreversible process taking place in an anisotropic crystal, such
that
J
α
=

β
L
αβ
X
β
, (2.29)
where fluxes and forces are vectors and L
αβ
is a tensor of order 2. The
Onsager–Casimir’s reciprocal relations write now as
L
αβ
= ±(L
βα
)
T
. (2.30)
Transformation properties of the reciprocal relations have been discussed by
Meixner (1943) and Coleman and Truesdell (1960).
At first sight, Onsager–Casimir’s reciprocal relations may appear as a
rather modest result. Their main merit is to have evidenced symmetry proper-
ties in coupled irreversible processes. As illustration, consider heat conduction
in an anisotropic crystal. The reciprocity relations imply that a temperature
gradient of 1

Cm
−1
along the x-direction will give raise to a heat flux in
the normal y-direction, which is the same as the heat flow generated along
the x-axis by a temperature gradient of 1

Cm
−1
along y. Another advan-
tage of the Onsager–Casimir’s reciprocal relations is that the measurement
(or the calculation) of a coefficient L
αβ
alleviates the repetition of the same
operation for the reciprocal coefficient L
βα
; this is important in practice as
the cross-coefficients are usually much smaller (of the order of 10
−3
to 10
−4
)
than the direct coupling coefficients, and therefore difficult to measure or
even to detect.
Although the proof of the Onsager–Casimir’s reciprocal relations was
achieved at the microscopic level of description and for small deviations of
fluctuations from equilibrium, these symmetry properties have been widely
applied in the treatment of coupled irreversible processes taking place at the
macroscopic scale even very far from equilibrium. It should also be kept in
mind that the validity of the reciprocity properties is secured as far as the
flux–force relations are linear, but that they are not of application in the
non-linear regime.
2.5 Stationary States
Stationary states play an important role in continuum physics; they are de-
fined by the property that the state variables, including the velocity, remain
unchanged in the course of time. For instance, if heat is supplied at one end
of a system and removed at the other end at the same rate, the temperature
at each point will not vary in time but will change from one position to the
2.5 Stationary States 51
other. Such a state cannot be confused with an equilibrium state, which is
characterized by a uniform temperature field, no heat flow, and a zero entropy
production. It is to be emphasized that the evolution of a system towards an
equilibrium state or a steady state is conditioned by the nature of the bound-
ary conditions. Since in a stationary state, entropy does not change in the
course of time, we can write in virtue of (2.7) that

¸
¸
¸
¸
dS
dt
¸
¸
¸
¸
out
+
¸
¸
¸
¸
dS
dt
¸
¸
¸
¸
in
+
d
i
S
dt
= 0, (2.31)
since the rate of entropy production is positive, it is clear that the entropy
delivered by the system to the external environment is larger than the entropy
that is entering. Using the vocabulary of engineers, the system degrades the
energy that it receives and this degradation is the price paid to maintain a
stationary state.
Stationary states are also characterized by interesting extremum principles
as demonstrated by Prigogine (1961): the most important is the principle of
minimum entropy production, which is discussed further. The importance of
variational principles has been recognized since the formulation of Hamilton’s
least action principle in mechanics stating that the average kinetic energy
less the average potential energy is minimum along the path of a particle
moving from one point to another. Quoting Euler, “since the construction
of the universe is the most perfect possible, being the handy work of an all-
wise Maker, nothing can be met in the world in which some minimum or
maximum property is not displayed”. It is indeed very attractive to believe
that a whole class of processes is governed by a single law of minimum or
maximum. However, Euler’s enthusiasm has to be moderated, as most of the
physical phenomena cannot be interpreted in terms of minima or maxima. In
equilibrium thermodynamics, maximization of entropy for isolated systems or
minimization of Gibbs’ free energy for systems at constant temperature and
pressure, for instance, provide important examples of variational principles.
Out of equilibrium, such variational formulations are much more limited.
For this reason, we pay here a special attention to the minimum entropy
production theorem, which is the best known among the few examples of
variational principles in non-equilibrium thermodynamics.
2.5.1 Minimum Entropy Production Principle
Consider a non-equilibrium process, for instance heat conduction or thermod-
iffusion taking place in a volume V at rest subjected to time-independent
constraints at its surface. The state variables a
1
, a
2
, . . . , a
n
are assumed to
obey conservation laws of the form
ρ
∂a
α
∂t
= −∇ J
α
(α = 1, 2, . . . , n), (2.32)
52 2 Classical Irreversible Thermodynamics
where ∂/∂t is the partial or Eulerian time derivative; processes of this kind,
characterized by absence of global velocity, are called purely dissipative. We
have seen in Sect. 2.4 that the total entropy P produced inside the system
can be written as
P =
_

α,β
L
αβ
X
α
X
β
dV. (2.33)
Since the thermodynamic forces take usually the form of gradients of intensive
variables
X
α
= ∇Γ
α
, (2.34)
where Γ
α
designates an intensive scalar variable, (2.33) becomes
P =
_

α,β
L
αβ
∇Γ
α
∇Γ
β
dV. (2.35)
We now wish to show that the entropy production is minimum in the sta-
tionary state. Taking the time derivative of (2.35) and supposing that the
phenomenological coefficients are constant and symmetric, one obtains
dP
dt
= 2
_

α,β
L
αβ
∇Γ
α

_
∂Γ
β
∂t
_
dV. (2.36)
After integration by parts and recalling that the boundary conditions are
time independent, it is found that
dP
dt
= −2
_

β
∂Γ
β
∂t

_

α
L
αβ
∇Γ
α
_
dV
= −2
_

β
∂Γ
β
∂t
∇ J
β
dV = 2
_
ρ

β
∂Γ
β
∂t
∂a
β
∂t
dV, (2.37)
wherein use has been made successively of the linear flux–force relations
(2.21) and the conservation law (2.32). In the stationary state, for which
∂α
β
/∂t = 0, one has
dP
dt
= 0. (2.38)
During the transient regime, (2.37) can be written as
dP
dt
=
_

α,β
∂a
α
∂Γ
β
∂Γ
α
∂t
∂Γ
β
∂t
dV, (2.39)
and since the ∂α
α
/∂Γ
β
terms (which represent, for instance, minus the heat
capacity or minus the coefficient of isothermal compressibility) are negative
quantities because of stability of equilibrium, one may conclude that
dP
dt
≤ 0. (2.40)
2.5 Stationary States 53
This result proves that the total entropy production P decreases in the course
of time and that it reaches its minimum value in the stationary state. An
important aside result is that stationary states with a minimum entropy pro-
duction are unconditionally stable. Indeed, after application of an arbitrary
disturbance in the stationary state, the system will move towards a transitory
regime with a greater entropy production. But as the latter can only decrease,
the system will go back to its stationary state, which is therefore referred to
as stable. It is also worth to mention that P, a positive definite functional
with a negative time derivative, provides an example of Lyapounov’s function
(Lyapounov 1966), whose occurrence is synonymous of stability, as discussed
in Chap. 6.
It should, however, be emphasized that the above conclusions are far
from being general, as their validity is subordinated to the observance of
the following requirements:
1. Time-independent boundary conditions
2. Linear phenomenological laws
3. Constant phenomenological coefficients
4. Symmetry of the phenomenological coefficients
In practical situations, it is frequent that at least one of the above restrictions
is not satisfied, so that the criterion of minimum entropy production is of weak
bearing. It follows also that most of the stationary states met in the nature
are not necessarily stable as confirmed by our everyday experience.
The result (2.38) can still be cast in the form of a variational principle
δP = δ
_
σ
s

α
, ∇Γ
α
, . . .) dV = 0, (2.41)
where the time derivative symbol d/dt has been replaced by the variational
symbol δ. Since the corresponding Euler–Lagrange equations are shown to
be the stationary balance relations, it turns out that the stationary state is
characterized by an extremum of the entropy production, truly a minimum,
as it can be proved that the second variation is positive definite δ
2
P > 0.
It should also be realized that the minimum entropy principle is not an
extra law coming in complement of the classical balance equations of mass,
momentum, and energy, but nothing else than a reformulation of these laws
in a condensed form, just like in classical mechanics, Hamilton’s principle is
a reformulation of Newton’s equations.
The search for variational principles in continuum physics has been a sub-
ject of continuous and intense activity (Glansdorff and Prigogine 1964, 1971;
Finlayson 1972; Lebon 1980). A wide spectrum of applications in macroscopic
physics, chemistry, engineering, ecology, and econophysics is discussed in
Sieniutycz and Farkas (2004). It should, however, be stressed that it is only in
exceptional cases that there exists a “true” variational principle for processes
that dissipate energy. Most of the principles that have been proposed refer
54 2 Classical Irreversible Thermodynamics
either to equilibrium situations, as the maximum entropy principle in equilib-
rium thermodynamics, the principle of virtual work in statics, the minimum
energy principle in elasticity, or to ideal reversible motions as the principle
of least action in rational mechanics or the minimum energy principle for
Eulerian fluids.
2.6 Applications to Heat Conduction, Mass Transport,
and Fluid Flows
To better understand and illustrate the general theory, we shall deal with
some applications, like heat conduction in a rigid body and matter diffusion
involving no coupling of different thermodynamic forces and fluid flow. More
complex processes involving coupling, like thermoelectricity, thermodiffusion
and diffusion through membranes are treated in Chap. 3. The selection of
these problems has been motivated by the desire to propose a pedagogical
approach and to cover situations frequently met in practical problems by
physicists, chemists and engineers. Chemistry will receive a special treatment
in Chap. 4 where we deal at length with chemical reactions and their coupling
with mass transport, a subject of utmost importance in biology. Despite its
success, CIT has been the subject of several limitations and criticisms, which
are discussed in Sect. 2.7 of the present chapter.
2.6.1 Heat Conduction in a Rigid Body
The problem consists in finding the temperature distribution in a rigid body
at rest, subject to arbitrary time-dependent boundary conditions on tem-
perature, or on the heat flux. Depending on the geometry and the physical
properties of the system and on the nature of the boundary conditions, a
wide variety of situations may arise, some of them being submitted as prob-
lems at the end of the chapter. For the sake of pedagogy, we follow the same
presentation as in Sect. 2.4.
Step 1. State variable(s)
Here we may select indifferently the specific internal energy u(r, t) or the
temperature field T(r, t), which should be preferred in practical applications.
Step 2. Evolution equation
In absence of source term, the evolution equation (2.18) for u(r, t) is simply
ρ
du
dt
= −∇ q, (2.42)
here d/dt reduces to the partial time derivative ∂/∂t as v = 0. Equation (2.42)
contains two unknown quantities, the heat flux q to be given by a constitutive
2.6 Applications to Heat Conduction, Mass Transport, and Fluid Flows 55
relation and the internal energy u(T) to be expressed by means of an equation
of state.
Step 3. Entropy production and second law
According to the second law, the rate of entropy production defined by
σ
s
= ρ
ds
dt
+∇ J
s
≥ 0, (2.43)
is positive definite. The expression of ds/dt is obtained from Gibbs’ equation
ds
dt
= T
−1
du
dt
, (2.44)
where du/dt is given by the energy balance equation (2.42). Substituting
(2.44) in (2.43) results in
σ
s
= q ∇T
−1
+∇ (J
s
−T
−1
q). (2.45)
Since σ
s
represents the rate of entropy production inside the body, its ex-
pression cannot contain a flux term like ∇ (J
s
−T
−1
q), which describes the
rate of exchange with the outside, as a consequence this term must be set
equal to zero so that
J
s
= T
−1
q, (2.46)
whereas (2.45) of σ
s
reduces to
σ
s
= q ∇T
−1
. (2.47)
This illustrates the general statement (2.20) that the entropy production is
a bilinear form in the force ∇T
−1
(the cause) and the flux of energy q (the
effect).
Step 4. Linear flux–force relation
The simplest way to ensure that σ
s
≥ 0 is to assume a linear relationship
between the heat flux and the temperature gradient; for isotropic media,
q = L
qq
(T)∇T
−1
, (2.48)
where L
qq
(T) is a scalar phenomenological coefficient depending generally on
the temperature. Defining the heat conductivity by λ(T) = L
qq
(T)/T
2
, the
flux–force relation (2.48) takes the more familiar form
q = −λ∇T, (2.49)
which is nothing else than the Fourier’s law stating that the heat flux is
proportional to the temperature gradient. We observe in passing that Curie’s
principle is satisfied as (2.49) is a relationship between flux and force of the
same tensor character, namely vectors. In an anisotropic crystal, Fourier’s
relation reads as
q = −λ ∇T, (2.50)
where the heat conductivity λ is now a tensor of order 2.
56 2 Classical Irreversible Thermodynamics
Step 5. Restriction on the sign of the transport coefficients
Substitution of (2.49) in (2.48) of the rate of entropy production yields
σ
s
=
1
λT
2
q q, (2.51)
and from the requirement that σ
s
≥ 0, it is inferred that λ ≥ 0. Roughly
speaking this means that in an isotropic medium, the heat flux takes place
in a direction opposite to the temperature gradient; therefore, heat will flow
spontaneously from high to low temperature, in agreement with our everyday
experience. In an anisotropic system, flux and force will generally be oriented
in different directions but the positiveness of tensor λ requires that the angle
between them cannot be smaller than π/2.
Step 6. Reciprocal relations
In the general case of an anisotropic medium, the flux–force relation is the
Fourier’s law expressed in the form (2.50). According to Onsager’s recipro-
cal relations, the second-order tensor λ is symmetric so that, in Cartesian
coordinates,
λ
ij
= λ
ji
, (2.52)
a result found to be experimentally satisfied in crystals wherein, however,
spatial symmetry may impose further symmetry relations. For instance, in
crystals pertaining to the hexagonal or the tetragonal class, spatial symme-
try requires that the conductivity tensor is skew-symmetric. By combining
this result with the symmetry property (2.52), it turns out that the elements
λ
ij
(i ,= j) are zero; it follows that for these classes of crystal, the application
of a temperature gradient in the x-direction cannot produce a heat flow in
the perpendicular y-direction. Very old experiences by Soret (1893) and Voigt
(1903) confirmed this result, which is presented as one of the confirmations
of the Onsager–Casimir’s reciprocal relations.
Step 7. The temperature equation
We now wish to calculate the temperature distribution in an isotropic rigid
solid as a function of time and space. The corresponding differential equation
is easily obtained by introducing Fourier’s law (2.49) in the energy balance
equation (2.42) and the result is
ρc
v
∂T
∂t
= ∇ (λ∇T), (2.53)
where use is made of the definition of the heat capacity c
v
= ∂u/∂T.
In the case of constant heat conductivity and heat capacity, and introduc-
ing the heat diffusivity defined by χ = λ/ρc
v
, (2.53) reads as
∂T
∂t
= χ

2
T
∂x
2
. (2.54)
Relation (2.53) is classified as a parabolic partial differential equation. In
Box 2.3 is presented the mathematical method of solution of this important
2.6 Applications to Heat Conduction, Mass Transport, and Fluid Flows 57
equation under given typical initial and boundary conditions for an infinite
one-dimensional rod. It is the same kind of equation that governs matter
diffusion, as shown in Sect. 2.6.2.
Box 2.3 Method of Solution of the Heat Diffusion Equation
A convenient method to solve (2.54) is to work in the Fourier space (k, t)
with k designating the wave number. It is interesting to recall that Fourier
devised originally the transform bearing his name to solve the heat diffusion
equation. Let us write T(x, t) as a Fourier integral of the form
T(x, t) =
_
+∞
−∞
T(k, t) exp(ikx) dk, (2.3.1)
with T(k, t) the Fourier transform. The initial and boundary conditions
are assumed to be given by T(x, 0) = g(x), T(±∞, t) = 0; introduction of
(2.3.1) in (2.54) leads to the ordinary differential equation
dT(k, t)
dt
= −k
2
χT(k, t), (2.3.2)
whose solution is directly given by
T(k, t) = T(k, 0) exp(−k
2
χt), (2.3.3)
where T(k, 0) is the Fourier transform of the initial temperature profile
T(k, 0) =
1

_
+∞
−∞
g(x

) exp(−ikx

) dx

. (2.3.4)
Substitution of (2.3.3) and (2.3.4) in (2.3.1) yields T(x, t) in terms of the
initial distribution g(x):
T(x, t) =
1

_
+∞
−∞
dk
_
+∞
−∞
g(x

) exp(−k
2
χt) exp[ik(x −x

)] dx

. (2.3.5)
By carrying the integration with respect to k, we obtain the final solution
in the form
T(x, t) =
1
(4πχt)
1/2
_
+∞
−∞
g(x

) exp[−(x −x

)
2
/4χt] dx

. (2.3.6)
When the initial temperature dependence corresponds to a local heating at
one particular point x
0
of the solid, namely
g(x) = g
0
δ(x −x
0
), (2.3.7)
where g
0
is an arbitrary constant proportional to the energy input at the
initial time, and δ(x −x
0
) the Dirac function, (2.3.6) will be given by
T(x, t) =
g
0
(4πχt)
1/2
exp[−(x −x
0
)
2
/4χt]. (2.3.8)
58 2 Classical Irreversible Thermodynamics
We deduce from (2.3.8) that the temperature diffuses in the whole rod from
the point x
0
over a distance proportional to (χt)
1/2
; as a consequence, large
values of the thermal diffusivity imply a rapid diffusion of temperature. Solu-
tion (2.3.8) exhibits also the property that, after application of a temperature
disturbance at a given point of the system, it will be experienced instanta-
neously everywhere in the whole body. Such a result is in contradiction with
the principle of causality, which demands that response will be felt after the
application of a cause. In the above example, cause and effect occur simul-
taneously. This property is typical of classical irreversible thermodynamics
but is not conceptually acceptable. This failure was one of the main motiva-
tions to propose another thermodynamic formalism currently known under
the name of Extended Irreversible Thermodynamics (see Chap. 7).
Box 2.4 Stationary States and Minimum Entropy Production
Principle
In the case of thermal conduction described by the phenomenological equa-
tion q = L
qq
∇T
−1
, the total entropy production P in the system, say a
one-dimensional rod of length l, is according to (2.47),
P =
_
l
0
σ
s
dx =
_
l
0
q(∂T
−1
/∂x) dx =
_
l
0
L
qq
(∂T
−1
/∂x)
2
dx. (2.4.1)
For a constant phenomenological coefficient L
qq
, its time derivative is
dP
dt
= 2
_
l
0
L
qq
∂T
−1
∂x

∂t
∂T
−1
∂x
dx. (2.4.2)
After integration by parts and imposing time-independent boundary con-
ditions, one has
dP
dt
= −2
_
l
0
∂T
−1
∂t

∂x
_
L
qq

∂x
T
−1
_
dx = 2
_
l
0
ρ
∂T
−1
∂t
∂u
∂t
dx, (2.4.3)
where use is made of the energy balance equation (2.42). Since ∂u/∂t =
c
v
∂T/∂t with c
v
> 0, it is finally found that
dP
dt
= −2
_
l
0
ρc
v
T
2
_
∂T
∂t
_
2
dx ≤ 0 (2.4.4)
is negative outside the stationary state and zero in the stationary state for
which ∂T/∂t = 0.
It is directly recognized that the total entropy production P is a Lya-
pounov function, as P > 0 together with dP/dt ≤ 0, and this ensures
that the stationary state is stable. It is, however, important to realize that
the phenomenological coefficient L
qq
is related to the heat conductivity by
L
qq
= λT
2
and therefore the validity of the minimum entropy production
2.6 Applications to Heat Conduction, Mass Transport, and Fluid Flows 59
principle is conditioned by the condition that λ varies like T
−2
; the princi-
ple is therefore not applicable to systems with a constant heat conductivity
or with an arbitrary dependence of λ with respect to the temperature. This
is the reason why the criterion of minimum entropy production remains an
exception.
2.6.2 Matter Diffusion Under Isothermal and Isobaric
Conditions
In this section, we briefly discuss the problem of isothermal and isobaric
diffusion of two non-viscous isotropic non-reacting fluids, in the absence of
external forces. There is no difficulty to generalize the forthcoming consid-
erations and results to the more general case of a n-component mixture of
viscous fluids.
Let ρ
1
and ρ
2
denote the densities (mass per unit volume) and c
1
(=
ρ
1
/ρ), c
2
(= ρ
2
/ρ) the mass fractions of the two constituents (c
1
+ c
2
= 1),
ρ = ρ
1

2
is the total mass density. Designating by v
k
(r, t) the local macro-
scopic velocity of substance k, the centre of mass or barycentric velocity is
given by
ρv = ρ
1
v
1

2
v
2
. (2.55)
The diffusion flux J
k
of substance k with respect to the centre of mass is
defined by
J
k
= ρ
k
(v
k
−v), (2.56)
and is expressed in kg m
−2
s
−1
; in some circumstances, it may be convenient
to replace in (2.56) the barycentric velocity by an arbitrary reference velocity
(de Groot and Mazur 1962). In virtue of the definition of the barycentric
velocity, it is directly seen that J
1
+J
2
= 0.
The analysis carried out in Sect. 2.6.1 for heat transport is easily repeated
for matter diffusion. In the present problem, the Gibbs’ equation takes the
form
T
ds
dt
=
du
dt
+p
dv
dt
−(¯ µ
1
− ¯ µ
2
)
dc
1
dt
, (2.57)
wherein ¯ µ
1
and ¯ µ
2
are the respective chemical potentials and where use has
been made of dc
2
= −dc
1
. Combining (2.57) with the balance equations
(2.16)–(2.18) written without source terms, it is easily shown (see Prob-
lem 2.6) that the entropy production is given by

s
= −J
1
∇(¯ µ
1
− ¯ µ
2
) ≥ 0. (2.58)
Within the hypothesis of linear flux–force relations, one obtains
J
1
= −L∇(¯ µ
1
− ¯ µ
2
) (L > 0), (2.59)
60 2 Classical Irreversible Thermodynamics
or, in virtue of Gibbs–Duhem’s relation c
1
d¯ µ
1
+c
2
d¯ µ
2
= c
1
∇¯ µ
1
+c
2
∇¯ µ
2
= 0,
J
1
= −
L
c
2
∇¯ µ
1
= −
L(∂¯ µ
1
/∂c
1
)
T,p
c
2
∇c
1
, (2.60)
wherein (∂¯ µ
1
/∂c
1
)
T,p
≥ 0 to satisfy the requirement of stability of equilib-
rium. After identifying the coefficient of ∇c
1
with ρD, where D (in m
2
s
−1
)
is the positive diffusion coefficient, one finds back the celebrated Fick’s law
J
1
= −ρD∇c
1
(D > 0). (2.61)
Elimination of J
1
between Fick’s law and the balance equation ρ dc
1
/dt =
−∇ J
1
leads to a parabolic differential equation in c
1
(r, t) similar to (2.54).
In the particular case of diffusion of N material particles located at a point
r = r
0
at time t
0
, the distribution of the mass fraction c
1
(r, t) obeys the
same law as (2.3.8), namely
c
1
(r, t) =
N
8(πDt)
3/2
exp[−(r −r
0
) (r −r
0
)/(4Dt)]. (2.62)
The characteristic displacement length is now given by l ∝

Dt, and this
result constitutes the main characteristic of diffusion phenomena, namely
that the mean distance of diffusion is proportional to the square root of the
time. Note that it is also usual to express Fick’s law in terms of the molar
concentration rather than the mass fraction; in such a case, Fick’s law takes
the form (2.2) instead of (2.61).
2.6.3 Hydrodynamics
As a further application, we will show that CIT provides a general frame-
work for the macroscopic description of hydrodynamics. The problem that
we wish to solve is the establishment of the complete set of partial differen-
tial equations governing the motion of a fluid. To be explicit, let us consider
the laminar motion (turbulence is excluded) of a one-constituent isotropic,
compressible and viscous fluid in presence of a temperature gradient, without
internal energy supply.
The set of basic variables giving a complete knowledge of the system in
space and time are the mass density ρ(r, t), the velocity v(r, t) and the
temperature T(r, t) fields.
These variables obey the classical balance equations of mass, momentum,
and energy already given in Sect. 2.4 but recalled here for the sake of clarity:
ρ

−1
dt
= ∇ v, (2.63)
ρ
dv
dt
= −∇ P+ρF, (2.64)
ρ
du
dt
= −∇ q −P : V. (2.65)
2.6 Applications to Heat Conduction, Mass Transport, and Fluid Flows 61
The quantity V is the symmetric part of the velocity gradient tensor; in
Cartesian coordinates, V
ij
=
1
2
(∂v
i
/∂x
j
+ ∂v
j
/∂x
i
). In fluid mechanics, the
symmetric pressure tensor P is usually split into a reversible hydrostatic
pressure pI and an irreversible viscous pressure P
v
in order that P = pI +
P
v
with I the identity tensor. The symmetric viscous pressure tensor P
v
may further be decomposed into a bulk part p
v
(=
1
3
tr P
v
) and a traceless
deviatoric part
0
P
v
so that P = p
v
I+
0
P
v
. The possibility and physical meaning
of an antisymmetric contribution to P
v
will be analysed in Problem 2.9.
Gathering all these results, the pressure tensor can be written as
P = (p +p
v
)I +
0
P
v
. (2.66)
Instead of the pressure tensor P, some authors prefer to use the stress tensor
σ, which is equal to minus the pressure tensor. All the results derived above
and the forthcoming remain valid by working with the stress tensor. The
evolution equations (2.63)–(2.65) constitute a set of five scalar relations with
16 unknown quantities namely, ρ, v, p, p
v
,
0
P
v
, q, u, and T. It is the aim
of the classical theory of irreversible processes to provide the eleven missing
equations. As usual, we start from Gibbs’ relation
ds
dt
= T
−1
du
dt
−pT
−1

−1
dt
, (2.67)
wherein du/dt and dρ
−1
/dt will be replaced, respectively, by expressions
(2.65) and (2.63). This yields the following balance equation for the specific
entropy s:
ρ
ds
dt
= −∇
q
T
+q ∇T
−1
−T
−1
p
v
∇ v −T
−1
0
P
v
:
0
V, (2.68)
where
0
V is the traceless part of tensor V. From (2.68), it is inferred that the
expressions of the entropy flux and the entropy production are given by
J
s
=
1
T
q, (2.69)
σ
s
= q ∇T
−1
−T
−1
p
v
∇ v −T
−1
0
P
v
:
0
V. (2.70)
Expression (2.70) is bilinear in the fluxes q, p
v
,
0
P
v
, and the forces ∇T
−1
,
T
−1
(∇ v), T
−1
0
V; by assuming linear relations between them and invoking
Curie’s law, one obtains the following set of phenomenological relations, also
called the transport equations:
q = L
qq
∇T
−1
≡ −λ∇T, (2.71)
p
v
= −l
vv
T
−1
∇ v ≡ −ζ∇ v, (2.72)
0
P
v
= −L
vv
T
−1
0
V ≡ −2η
0
V. (2.73)
62 2 Classical Irreversible Thermodynamics
We recognize in (2.71) the Fourier’s law by identifying the phenomenolog-
ical coefficient L
qq
/T
2
with the heat conductivity λ; (2.72) is the Stokes’
relation if l
vv
/T is identified with the bulk viscosity ζ, and finally (2.73) is
the Newton’s law of hydrodynamics when L
vv
/T is put equal to 2η, with η
the dynamic shear viscosity. It follows from (2.71)–(2.73) that the fluid will be
set instantaneously in motion as soon as it is submitted to a force, in contrast
with elastic solids, which may stay at rest even when subject to stresses. At
equilibrium, both members of equations (2.71)–(2.73) vanish identically as it
should. The phenomenological coefficients λ, ζ, and η depend generally on ρ
and T but in most practical situations the dependence with respect to ρ is
negligible. In the case of incompressible fluids for which ∇ v = 0, the viscous
pressure p
v
is zero and the bulk viscosity does not play any role. However,
in compressible fluids like dense gases and bubbly liquids, the bulk viscosity
is by no means negligible. The bulk viscosity vanishes in the case of perfect
gases; in real gases, it is far from being negligible and the ratio ζ/η between
bulk and shear viscosity may even be of the order of hundred.
After substitution of the transport equations (2.71)–(2.73) in (2.70) of the
entropy production, one obtains
σ
s
=
λ
T
2
(∇T)
2
+
ζ
T
(∇ v)
2
+

T
(
0
V :
0
V) ≥ 0. (2.74)
Positiveness of σ
s
requires that
λ > 0, ζ > 0, η > 0. (2.75)
Negative values of these transport coefficients will therefore be in contradic-
tion with the second law of thermodynamics.
By combining the nine transport equations (2.71)–(2.73) with the five
evolution equations (2.63)–(2.65), one is faced with a set of 14 equations
with 16 unknowns. The missing relations are provided by Gibbs’ equation
(2.67) from which are supplemented the two equations of state T = T(u, ρ),
p = p(u, ρ), or solving with respect to u,
u = u(T, ρ), p = p(T, ρ). (2.76)
These equations, together with the evolution and transport relations describe
completely the behaviour of the one-component isotropic fluid after that
boundary and initial conditions have been specified. The state equations
(2.76) have a status completely different from the phenomenological equa-
tions (2.71)–(2.73) because, in contrast with the latter, they do not vanish
identically at equilibrium. By substitution of the phenomenological laws of
Stokes (2.72) and Newton (2.73) in the momentum balance equation (2.64),
one finds back the well-known Navier–Stokes’ equation when it is assumed
that the viscosity coefficients are constant:
ρ
dv
dt
= −∇p + 2η∇
2
v +
_
2
3
η +ζ
_
∇(∇ v) +ρF. (2.77)
2.7 Limitations of the Classical Theory of Irreversible Thermodynamics 63
In the case of an isothermal and incompressible fluid, which is the most
customary situation met in hydrodynamics, the behaviour of the fluid is com-
pletely described by (2.77) and ∇ v = 0. For more general situations like
multi-component mixtures with diffusion, the equations become much more
complicated but the way to derive them follows the same systematic proce-
dure as above. It should, however, be stressed that the results of this section
are only applicable to the study of laminar flows of some particular class of
fluids, namely these described by the linear flux–force relations of Fourier,
Stokes, and Newton. The above formalism is not adequate for non-Newtonian
fluids or polymeric solutions and cannot cope with turbulent motions. We
note finally that the basic relations obtained so far, like for instance the
Navier–Stokes’ equation (2.77), are parabolic partial differential equations,
which means that disturbances will propagate through the fluid with an in-
finite velocity.
2.7 Limitations of the Classical Theory
of Irreversible Thermodynamics
Despite its numerous successes, CIT has raised several questions and some
shortcomings have been pointed out. Let us comment briefly about the most
frequent criticisms:
1. The theory is not applicable to irreversible processes described by non-
linear phenomenological equations. Many actual processes, like chemical
reactions or non-Newtonian flows, are indeed characterized by non-linear
phenomenological relations and are therefore outside the scope of classical
irreversible thermodynamics. It should, however, be noted that linearity
between fluxes and forces does not require that the phenomenological co-
efficients, like the heat conductivity and the viscosity coefficients, are con-
stant; they may indeed depend in particular on the temperature and the
pressure with the consequence that the corresponding evolution equations,
like Navier–Stokes’ equation, may be non-linear.
2. The cornerstone of the theory is the local equilibrium hypothesis, accord-
ingly the thermodynamic state variables are the same as in equilibrium
and Gibbs’ equation, which plays a fundamental role in CIT, remains lo-
cally valid, i.e. at each position in space and each instant of time. The
advantage of this assumption is that entropy has a precise meaning, even
in non-equilibrium states, and that it leads to an explicit expression for
the entropy production from which are inferred appropriate forms for the
constitutive relations. However, it is conceivable that other variables, not
found at equilibrium, are able to influence the process. This is, for instance,
the case of polymers of long molecular chains in which configuration in-
fluences considerably their behaviour. Other examples are superfluids and
superconductors whose peculiar properties ask for the introduction of extra
64 2 Classical Irreversible Thermodynamics
variables. As the local equilibrium hypothesis implies large time and space
scales, CIT is not appropriate for describing high-frequency phenomena
as ultrasound propagation or nuclear collisions and short-wavelength sys-
tems and processes, like nano-structures, shock waves or light and neutron
scattering.
Another important point to notice is that, when referred to general-
ized hydrodynamics, the phenomenological coefficients are frequency and
wavelength dependent. This is at variance with the local equilibrium hy-
pothesis, implying that these coefficients are frequency and wavelength
independent.
3. By introduction of the Fourier’s law with constant heat conductivity in
the energy balance, one obtains an equation for the temperature, which
is of the diffusion type; see for instance (2.54). From the mathematical
point of view, this is a parabolic partial differential equation, which im-
plies that after application of a disturbance, the latter will propagate at
infinite velocity across the body so that it will be felt instantaneously and
everywhere in the whole body. The same remark is applicable to the other
equations of the theory and in particular to Navier–Stokes’ equation. The
problem of propagation of signals with an infinite velocity is conceptually
a major inconvenient, which is not acceptable from a purely physical point
of view. However it is not a tragedy in most practical situations because
their characteristic time is much longer than the transit time of the signals
transporting the thermodynamic information.
4. Another important subject of controversy concerns the validity of the
Onsager–Casimir’s reciprocal relations, which were derived from the con-
dition of microscopic reversibility. Although the reciprocity property was
originally shown by Onsager to be applicable at the microscopic level for
particles in situations close to equilibrium, it is current to extrapolate
this result at the macroscopic level for continuum media driven far from
equilibrium. There is clearly no theoretical argument supporting such an
extrapolation. Onsager’s demonstration itself has been the subject of acrid
criticisms because it is based on an important assumption with regard to
the regression of the fluctuating thermodynamic variables. Indeed, it lies
on the hypothesis that the rate of decay of the fluctuations takes place
according to the same law as the macroscopic linear flux–force relations.
Moreover, in the light of the demonstration given by Onsager, the thermo-
dynamic fluxes are defined as time derivatives of extensive thermodynamic
variables and the forces are the derivatives of the entropy with respect to
the same state variables. Such conditions are certainly not respected by
thermodynamic fluxes like the heat flux or the pressure tensor or ther-
modynamic forces as the temperature gradient or the velocity gradient
tensor.
5. As shown explicitly in Box 2.2, the Curie law stating that in isotropic ma-
terials, fluxes couple only with forces of the same tensorial order, is directly
derivable from algebra. It is also important to realize that this exclusion
2.8 Problems 65
law is not valid outside the linear regime even for isotropic bodies. As
observed by Truesdell (1984), these results were well known even before
Curie was born. Moreover, when one goes back to Curie’s original papers
on symmetry considerations, there is nothing, even vaguely mentioned,
which is reminding of the law bearing his name in classical irreversible
thermodynamics. In anisotropic systems, the coupling between terms of
different tensorial order, as for instance chemical reactions (scalar) and
mass transport (vectorial), may provide very important effects, as shown
in Chap. 4.
6. Another point of debate is the question of how the fluxes and forces are
to be selected from the expression of the entropy production. It is gener-
ally admitted that the thermodynamic fluxes are the physical fluxes that
appear in the balance laws of mass, momentum, and energy, and that the
forces are the conjugated terms in the bilinear expression of the entropy
production. Another suggestion is to identify the forces as the gradient
of intensive variables, as temperature and velocity, and to select as fluxes
the conjugate terms in the entropy production. Unfortunately, such defin-
itions are not applicable to chemical reactions. Another school of thought
claims that it is indifferent how the fluxes and forces are selected, at the
condition that non-singular linear transformations leave invariant the en-
tropy production expression. However, as pointed out by Coleman and
Truesdell (1960), one should be careful in the identification of fluxes and
forces, as a wrong selection could destroy the reciprocal property of the
phenomenological coefficients. Clearly, the controversy about how fluxes
and forces are chosen is not closed but, in our opinion, this is not the more
fundamental question about CIT, all the more as the discussions about
this point have been sterile.
2.8 Problems
2.1. Heat transport and entropy production. Two heat reservoirs at tempera-
ture T
1
and T
2
are connected by a rigid heat-conducting one-dimensional rod.
Find the longitudinal temperature profiles in the rod if the heat conductivity
is (a) constant and (b) proportional to T
α
, where α is a constant. Calculate
the entropy produced along the rod per unit time.
2.2. Fourier’s and Newton’s laws. Consider a rigid cylindrical heat-conducting
rod. The heat transfer along the rod is described by means of Fourier’s
law (q = −λ∇T), whereas the heat exchange with the environment is de-
scribed by Newton’s cooling law, expressing that the heat flow is propor-
tional to the temperature difference between the rod and the environment
(q n = −α(T −T
env
)), where n is the unit vector normal to the boundaries.
(a) Derive the equation describing the temperature distribution along the
66 2 Classical Irreversible Thermodynamics
rod. (b) Determine the temperature profile along a very long rod if its hotter
boundary is kept at T
1
and the environment’s temperature T
env
is constant.
2.3. Temperature profile. A cylindrical tube of radius R and length L is im-
mersed in an environment at temperature T
0
. A fluid flows inside the tube at
a flow rate Q; the temperature of the fluid when it enters the tube is T
1
. Find
the temperature T
2
of the outgoing fluid at the end of the tube as a function
of the flow rate Q in terms of its specific heat per unit mass, its density and
its thermal conductivity. The heat transfer coefficient across the lateral walls
of the tube is a.
2.4. Radiative heat transfer. Assume that heat exchange between a spher-
ical body of radius R and its environment occurs through electromagnetic
radiation and obeys Stefan–Boltzmann’s law. Accordingly, the quantity of
heat emitted per unit area and unit time by a body at temperature T is
q = εσT
4
, where σ is the Stefan–Boltzmann constant and ε the emissivity
of the system. (a) Establish the differential equation describing the cooling
of the sphere when its surface is at temperature T and the environment at
temperature T
env
. (b) Find the corresponding entropy production per unit
time. (c) Solve the problem when the difference T −T
env
is small compared
to T
env
. (d) How will the solution be modified if the radius of the sphere is
10 times larger?
2.5. The age of the Earth. A semi-infinite rigid solid of heat diffusivity χ,
initially at temperature T
10
, is suddenly put into contact with another semi-
infinite solid at fixed temperature T
2
. Assume that heat exchange is described
by Fourier’s law. (a) Show that the temperature profile as a function of time
and position is
θ(x, t) =
θ
0

πχt
_
x
−∞
dx

exp
_

(x −x

)
2
4χt
_
,
where θ = T(x, t) − T
2
and θ
0
≡ T
10
− T
2
. (b) Verify that the temperature
gradient at the surface is
_
∂θ
∂x
_
x=0
= −
θ
0

πχt
.
(c) At the end of the nineteenth century, lord Kelvin evaluated the age of
the Earth from an analysis of the terrestrial temperature gradient near the
surface of the Earth, found to be equal to −3 10
−2◦
Cm
−1
. By taking
χ ≈ 310
5
m
2
s
−1
and θ
0
≈ 3, 800

C, estimate the age of the Earth according
to this model. The result found is much less than the actual estimation of
4.5 billion years (radioactivity, which was not known at Kelvin’s time, will
considerably influence the result).
2.6. Matter diffusion. Derive the expressions of the entropy flux and the
entropy production for the problem of isothermal and isobaric diffusion in a
binary mixture of non-viscous fluids, in absence of external body forces (see
Sect. 2.6.2).
2.8 Problems 67
2.7. Concentration profile. A perfectly absorbing sphere of radius R is im-
mersed in a solvent–solute solution with homogeneous concentration c
0
. The
diffusion coefficient of the solute is D. Since the sphere is perfectly absorbing,
the concentration of solute on its surface is always zero. (a) Determine the
concentration profile as a function of the radial distance. (b) Show that the
amount of solute crossing the surface of the sphere per unit time is given
by I = 4πDRc
0
. Hint: In spherical symmetry, the diffusion equation has the
form
∂c
∂t
=
D
r
2

∂r
_
r
2
∂c
∂r
_
.
2.8. The Einstein relation. A dilute suspension of small particles in a viscous
fluid at homogeneous temperature T is under the action of the gravitational
field. The friction coefficient of the particles with respect to the fluid is a (a =
6πηr for spherical particles of radius r in a solvent with viscosity η). Owing to
gravity, the particles have a sedimentation velocity v
sed
= m

g/a, with m

the
mass of one particle minus the mass of the fluid displaced by it (Archimede’s
principle); the corresponding sedimentation flux is J
sed
= nv
sed
, with n the
number of particles per unit volume. Against the sedimentation flux is acting
a diffusion flux J
dif
= −D∂n/∂z, D being the diffusion coefficient. (a) Find
the vertical profile of n(z) in equilibrium. (b) Compare this expression with
Boltzmann’s result
n(z) = n(0) exp[−(m

gz/k
B
T)],
where k
B
is the Boltzmann constant, and demonstrate Einstein’s relation
D = k
B
T/a.
2.9. Micropolar fluids. In some fluids (composed of elongated particles or
rough spheres) the pressure tensor is non-symmetric. Its antisymmetric part
is related to the rate of variation of an intrinsic angular momentum, and
therefore, it contributes to the balance equation of angular momentum. The
antisymmetric part of the tensor is usually related to an axial vector P
va
,
whose components are defined as P
va
1
= P
va
23
, P
va
2
= P
va
31
, P
va
3
= P
va
12
. The
balance equation for the internal angular momentum is
ρj ˙ ω +∇ Q = −2P
va
,
with j the microinertia per unit mass of the fluid, w the angular velocity, and
Q the flux of the intrinsic angular momentum, which is usually neglected.
(a) Prove that the corresponding entropy production is
σ
s
= −T
−1
p
v
∇ v −T
−1
0
P
vs
: (
0
∇v)
s
−T
−1
P
va
(∇v −2ω) +q ∇T
−1
,
where
0
P
vs
is the symmetric part of P
v
. Hint: Note that ∂s/∂ω = −ρjT
−1
ω.
(b) Show that the constitutive equation for P
va
may be given by
68 2 Classical Irreversible Thermodynamics
P
va
= −η
r
(∇v −2ω),
where η
r
is the so-called rotational viscosity; explain why there is no coupling
between P
va
and q (see, for instance, Snider and Lewchuk 1967; Rub´ı and
Casas-V´azquez 1980).
2.10. Flux–force relations. It is known that by writing a linear relation of the
form J
α
=

β
L
αβ
X
β
between n-independent fluxes J
α
(α = 1, . . . , n) and
n-independent thermodynamic forces X
β
, the matrix L
αβ
of phenomenologi-
cal coefficients is symmetric, according to Onsager. (a) Show that the matrix
L
αβ
is still symmetric when not all the fluxes are independent but one of
them is a linear combination of the other ones, i.e.

n
α=1
a
α
J
α
= 0. (b) Does
this conclusion remain fulfilled when instead of a linear relation between the
fluxes, there exists a linear relation

n
α=1
a
α
X
α
= 0 between the forces?
2.11. Minimum entropy production principle. Consider a one-dimensional
rigid heat conductor with fixed temperatures at its ends. Show that the
Euler–Lagrange equation corresponding to the principle of minimum entropy
production
δP(T
−1
) = δ
_
q(∂T
−1
/∂x) dx = 0
is the steady energy balance ∂q/∂x = 0 at the condition to write Fourier’s
law in the form q = L∂T
−1
/∂x with L = constant. Show that δ
2
P ≥ 0; this
result expresses that the Euler–Lagrange equation is a necessary condition
for P to be a minimum. Is the principle still fulfilled when Fourier’s law is
written in the more classical form q = −λ∂T/∂x with λ = constant?
2.12. Variational principles. Prigogine’s minimum entropy production is not
the only variational principle in linear non-equilibrium thermodynamics.
Other principles have been proposed, amongst others, by Onsager (1931) and
Gyarmati (1970). For instance, consider the entropy production P(X, J) ≡

i
X
i
J
i
and the dissipation functions
Φ(J, J) ≡
1
2

i,j
R
ij
J
i
J
j
and Ψ(X, X) ≡
1
2

i,j
L
ij
X
i
X
j
,
X being the thermodynamic forces and J the conjugated fluxes. Show that if
the matrices L
ij
, R
ij
, and L
ij
are constant and symmetric, the two following
variational principles are valid (1) for prescribed forces X and varying fluxes
J, P(X, J) − Φ(J, J) is maximum implies that X
k
=

i
R
ki
J
i
and (2) for
prescribed fluxes J and varying forces X, P(X, J) − Ψ(X, X) is maximum
implies that J
k
=

i
L
ki
X
i
. (The reader interested in these topics is referred
to Ichiyanagi 1994.)
Chapter 3
Coupled Transport Phenomena
Thermoelectricity, Thermodiffusion, Membranes
The analysis of coupled processes is one of the most outstanding aspects of
the classical theory of non-equilibrium thermodynamics. To emphasize this
feature, this chapter is dedicated to the study of coupled transport processes,
involving two fluxes and two thermodynamic forces. Emphasis is put on some
practical applications in thermoelectricity (coupling of heat and electric-
ity), thermodiffusion (coupling of heat and mass motion) and transport of
matter through membranes. In the precedent chapter, we have already out-
lined, in general terms, the importance of the Onsager’s reciprocal relations
ruling the coefficients that describe such couplings. Here, we will discuss di-
rect applications.
Coupled transport processes find a natural theoretical framework in non-
equilibrium thermodynamics. However, they were discovered well before the
developments of these theories during the early nineteenth century. Indeed,
thermoelectric effects were observed by Seebeck and Peltier, respectively, in
1821 and 1835, even earlier than the quantitative formulation of Ohm’s law in
1855. The first results on thermodiffusion were obtained by Ludwig in 1856,
Dufour’s effect was discovered in 1872, and Soret carried out his first system-
atic researches in 1879. Although these topics are rather old and classical,
from the practical point of view they have entered recently in a completely
new perspective, because of the recent developments in materials sciences,
high-power lasers and renewed interest in optimization of energy generation.
We will outline here these new features.
Transport properties play a central role in materials sciences. For instance,
a problem of acute interest is to obtain superconductor materials with high
critical temperatures. Another domain of intense activity is the optimal ap-
plication of thermoelectricity to convert waste heat to usable electricity or to
develop cooling devices; this requires materials with high Seebeck coefficient,
high electrical conductivity and low thermal conductivity. This is not an easy
task because usually a high electrical conductivity implies also a high ther-
mal conductivity. Recently, research on thermoelectric materials has known
an extraordinary burst: it is frequent to find semiconductors and superlattice
69
70 3 Coupled Transport Phenomena
devices whose figure of merit (a combination of Seebeck coefficient, electrical
and thermal conductivities directly related to the efficiency of energy con-
version) has increased by more than one order of magnitude since the end of
the twentieth century (Mahan et al. 1997). This was in particular achieved
in superlattices in which one dimension is smaller than the mean free path of
phonons but larger than that of electrons. More generally, it is desirable to
combine a macroscopic approach of transport phenomena, to design the most
efficient devices, with a microscopic understanding of the basic phenomena.
In parallel with thermoelectricity, practical applications of thermodiffusion
have also known a renewal of interest during the last decades with the possi-
bility of producing much localized heating of fluids by means of high-power
and extremely focused lasers. High temperature gradients will promote the
separation of different constituents in mixtures.
Finally, since the problem of transport across membranes remains a subject
of constant and wide interest, mainly in life sciences, we have reserved one
section of this chapter to its analysis.
3.1 Electrical Conduction
The results presented in Chap. 2 are directly applicable to the problem of
motion of electric charges in a conductor. First, we briefly study the flow
of electrical carriers in a rigid conductor at rest and in absence of thermal
effects. The basic variables are (1) z, the electrical charge of the carrier, say
electrons, per unit mass (z should not be confused with the charge z
e
per
unit mass of the electrons to which it is related by z = c
e
z
e
, where c
e
is the
mass fraction of electrons) and (2) u the internal energy per unit mass. The
evolution equations are the law of conservation of charges and the internal
energy balance, i.e.
ρ
dz
dt
= −∇ i , (3.1)
ρ
du
dt
= E i , (3.2)
where E is the electric field and i the density of conductive current (current
intensity per unit transversal area); in whole generality, the latter is related
to the density I of total current I by I = i +ρzv, where ρzv is the convective
current. In the present problem, one has v = 0 so that I = i while d/dt will
reduce to the partial time derivative.
The Gibbs’ equation (2.4) simplifies as T ds = du − µ
e
dc
e
, with µ
e
the
chemical potential of the electrons. In terms of the time derivative and the
charge density z, Gibbs’ relation takes the form
T
ds
dt
=
du
dt

µ
e
z
e
dz
dt
. (3.3)
3.1 Electrical Conduction 71
Substitution of (3.1) and (3.2) in (3.3) yields
ρ
ds
dt
= ∇
_
µ
e
Tz
e
i
_
+
1
T
_
E −∇
µ
e
z
e
_
i , (3.4)
from which it is deduced that the entropy flux and the entropy production
are given, respectively, by
J
s
=
µ
e
Tz
e
i , (3.5)

s
=
_
E −∇
µ
e
z
e
_
i ≥ 0. (3.6)
The last result suggests writing the following linear flux–force relation
E −∇
µ
e
z
e
= ri , (3.7)
or, for negligible values of ∇(µ
e
/z
e
),
E = ri , (3.8)
which is Ohm’s law with r the electrical resistivity. After introducing Ohm’s
law in (3.6), one obtains Tσ
s
= ri
2
, which represents Joule’s heating and
from which is inferred that the electrical resistivity is a positive quantity.
In the presence of an external magnetic field H, interesting coupled effects
appear and the resistivity is no longer a scalar, but a second-order tensor
because of the anisotropy introduced by H. In this case, the resistivity de-
pends on the magnetic field and the Onsager–Casimir’s reciprocity relations
state that r
ij
(H) = r
ji
(−H). If the external magnetic field is applied in
the z-direction with all currents and gradients parallel to the x–y plane, the
resistivity tensor r will take the form (Problem 3.2)
r =



r
11
(H) r
12
(H) 0
−r
12
(H) r
22
(H) 0
0 0 r
33
(H)


⎠, (3.9)
where H is the component in the z-direction. It follows that an electric field
in the x-direction will produce a current not only in the parallel direction but
also in the normal y-direction, a phenomenon known as the isothermal Hall
effect. This effect is easily understandable, as the external magnetic field will
deflect the charge carriers along a curved trajectory. The discovery, at the
beginning of the 1980s, of subtle phenomena such as quantum Hall effect,
boosted an enormous impetus on this topic. The dependence of the diagonal
components on the magnetic field, known as magneto-resistance, has led to
important applications in computer technology. These topics, however, are
beyond the scope of purely macroscopic considerations.
72 3 Coupled Transport Phenomena
3.2 Thermoelectric Effects
The coupling between temperature gradient and electrical potential gradient
gives rise to the thermoelectric effects of Seebeck, Peltier, and Thomson. Here
we present them in the framework of classical irreversible thermodynamics,
outline their main physical features and, as a practical application, we will
determine the efficiency of thermoelectric generators.
3.2.1 Phenomenological Laws
In presence of both thermal and electrical effects, the law of conservation
of charge (3.1) remains unchanged while the energy balance equation (3.2)
contains an additional term and reads as
ρ
du
dt
= −∇ q +E i . (3.10)
Combining (3.1) and (3.10) with the Gibbs’ equation (3.3), it is a simple
exercise (see Problem 3.1) to prove that the entropy flux and the rate of
dissipation per unit volume Tσ
s
are, respectively, given by
J
s
=
q
T

µ
e
Tz
e
i , (3.11)

s
= −J
s
∇T +
_
E −∇
_
µ
e
z
e
__
i . (3.12)
Restricting to isotropic media, the corresponding phenomenological equations
are
J
s
= −L
11
∇T +L
12
_
E −∇
_
µ
e
z
e
__
, (3.13)
i = −L
21
∇T +L
22
_
E −∇
_
µ
e
z
e
__
, (3.14)
with L
12
= L
21
in virtue of Onsager’s reciprocal relations. For practical
reasons, it is convenient to resolve (3.13) and (3.14) with respect to J
s
and
E −∇(µ
e
/z
e
) and to introduce the following phenomenological coefficients:
λ = T(L
11
−L
12
L
21
/L
22
) (heat conductivity),
π = T(L
12
/L
22
) (Peltier coefficient),
ε = L
21
/L
22
(Seebeck coefficient),
r = 1/L
22
(electrical resistivity).
3.2 Thermoelectric Effects 73
With this notation, (3.13) and (3.14) take the more familiar forms
q = −λ∇T +
_
π +
µ
e
z
e
_
i , (3.15)
E −∇
_
µ
e
z
e
_
i = ε∇T +ri , (3.16)
with, as a consequence of the symmetry property L
12
= L
21
,
π = εT. (3.17)
This result is well known in thermoelectricity as the second Kelvin rela-
tion and has been confirmed experimentally and theoretically in transport
theories.
In most applications, it is permitted to omit the contribution of the terms
in µ
e
/z
e
in relations (3.15) and (3.16), which receive then a simple interpre-
tation. It follows from (3.15) that, even in absence of a temperature gradient,
a heat flux may be generated due to the presence of an electrical current: this
is called the Peltier’s effect. The coefficient π = ±([q[/[i [)
∆T=0
is a measure
of the quantity of heat absorbed or rejected at the junction of two conductors
of different materials kept at uniform temperature and crossed by an electric
current of unit density (see Box 3.2).
Expression (3.16) exhibits the property that, even in absence of an electric
current, an electrical potential E = −∇φ can be created due solely to a
temperature difference, this is known as Seebeck’s effect: the coefficient ε =
±(∆φ/∆T)
i =0
is the Seebeck or thermoelectric power coefficient, it measures
the electrical potential produced by a unit temperature difference, in absence
of electric current. The Seebeck’s effect finds a natural application in the
construction of thermocouples (see Box 3.1).
Box 3.1 The Seebeck’s Effect and Thermoelectric Power
The most relevant application of the Seebeck’s effect is the thermocou-
ple, which is frequently used for temperature measurements. The latter is
constituted by two different materials (usually metals or alloys) forming
a circuit with some points at different temperatures. Consider the situa-
tion presented in Fig. 3.1, where the two junctions between the materials
A and B are at different temperatures T
1
and T
2
, respectively. A voltmeter
is inserted inside the system, at a position where temperature is T
0
. In
absence of electric current, the voltmeter indicates a value ∆Φ for the elec-
tric potential difference, which is related to ∆T ≡ T
2
− T
1
, by means of
∆Φ =
_
T
2
T
1

A
−ε
B
) dT. Assuming that ε
A
− ε
B
is known and constant,
one has ∆Φ = (ε
A
− ε
B
)∆T, and the measurement of ∆Φ will allow us
to determine the value of ∆T. Usually, one of the two temperatures is a
reference temperature (for instance, the melting point of ice) and the other
one is the temperature to be measured.
74 3 Coupled Transport Phenomena
Fig. 3.1 Thermocouple
The thermoelectric power of the thermocouple is defined as the dif-
ference between the absolute thermoelectric powers of the two materials,
namely ε
AB
= ε
B
− ε
A
. Intuitively, the thermoelectric power reflects the
tendency of carriers to diffuse from the hot end (characterized by a higher
thermal energy whence higher average kinetic energy of the carriers) to the
cold end, carrying with them their electric charge. The sign of the See-
beck coefficient ε is taken as positive when the carriers are positive charges
(as for instance holes in p-doped semiconductors) and negative otherwise
(as in most metals or n-doped semiconductors). This rule is not respected
in some metals, like copper or silver, which are characterized by a positive
thermoelectric power. The usual values of ε are of the order of 10 µVK
−1
for metals and 100 µVK
−1
for semiconductors. Typical examples of thermo-
couples are those formed by the junctions of copper/constantan (a copper–
nickel alloy), nickel–chromium alloy/constantan, or platinum–rhodium
alloy/platinum.
Another interesting application of the Seebeck’s effect is thermoelectric
generation, namely electrical power generated from a temperature differ-
ence; the discussion of this subject will be postponed later.
Box 3.2 The Peltier’s Effect and Thermoelectric Cooling Systems
The Peltier’s effect describes the absorption or emission of heat accompa-
nying the flow of an electric current across the junctions of two materials
A and B, under isothermal conditions (see Fig. 3.2). Assume, for instance,
that the circulation of the current from A to B produces a cooling of the
junction between them. The effect is reversed by reversing the direction of
the electric current.
To interpret this phenomenon from a microscopic perspective is easy.
If in material A the electrons are moving at an energy level lower than in
material B, the transition of one electron from A to B will require the supply
of some energy from the junction and therefore the passage of current is
accompanied by cooling the right junction. The reverse is true by transiting
from B to A: some quantity of energy is released at the left junction which
3.2 Thermoelectric Effects 75
Fig. 3.2 The Peltier’s effect
is heated. An example is that of electrons flowing from a low energy level in
p-type semiconductors to a higher-energy level in n-type semiconductors.
Materials frequently used are bismuth telluride (Bi
2
Te
3
) heavily doped and,
more recently, the layered oxide Na
x
Co
2
O
4
.
A practical application of the Peltier’s effect is the thermoelectric re-
frigerator. If the junction A → B is cooled, when electric current flows
in a given direction, it will be put inside the container to be refrigerated,
whereas the junction B → A, delivering heat, is left outside. The effect of
the passage of a current is thus to extract heat from the container and to
deliver it to the external world. When current flows in the opposite direc-
tion, the reverse process occurs, namely extracting heat from the external
world and heating the container. The analogy with usual refrigerators or
heat pumps based on Carnot cycle is the following: the junction absorbing
heat is analogous to the expansion of refrigerant in contact with the cold
reservoir whereas the junction delivering heat is the analogous of the com-
pression of the refrigerant in contact with the hot reservoir. Thermoelectric
coolers have the advantage of not having moving parts, since there is neither
compression nor expansion of the refrigerant, nor phase change. They are
reliable, do not need maintenance and can be given a very compact form.
Nevertheless, they present the disadvantage to consume more energy and
are therefore less efficient than usual coolers.
A third thermoelectric coupling of interest is Thomson’s heat. This
effect occurs from the simultaneous presence of a temperature gradient and
an electric current, and is the result of the coupling of both the Peltier’s
and Seebeck’s effects. Substituting (3.15) and (3.16) in the energy balance
equation (3.10) and taking into account of the conservation of total charge
(∇ i = 0) and Kelvin’s second relation, it is found that
ρ
du
dt
= ∇ (λ∇T) +ri
2
−i (∇π)
T
+
_
π
T

∂π
∂T
_
i ∇T, (3.18)
76 3 Coupled Transport Phenomena
where subscript T means that the temperature is fixed. If the temperature
remains uniform, only the Joule effect ri
2
and the Peltier’s effect i (∇π)
T
contribute to the dissipated heat. Now, in presence of a temperature gradi-
ent, two supplementary terms occur in (3.18): heat conduction plus a term
resulting from the coupling i ∇T of the electric current and the tempera-
ture gradient and called the Thomson’s heat. The latter may be positive or
negative in contrast with Joule heating, which is invariably positive.
The coefficient
σ
Th

π
T

∂π
∂T
(3.19)
is the so-called Thomson coefficient, which, in virtue of π = εT can still be
cast in the form
σ
Th
= −T
∂ε
∂T
. (3.20)
This result is referred to as the first Kelvin relation and exhibits the strong
connection between the various thermoelectric effects.
3.2.2 Efficiency of Thermoelectric Generators
We have shortly presented in Boxes 3.1 and 3.2 some practical applications
of thermoelectric materials. Motivated by the developments of research on
new materials (Mahan et al. 1997), let us calculate the efficiency of thermo-
electric conversion. Consider a single one-dimensional thermoelectric element
of length l under steady conditions. The hot side is at a temperature T
h
(as-
sumed to be the upper side, at y = l), and the cold side (the lower side, at
y = 0) is at temperature T
c
. An electric current i and a quantity of heat
˙
Q enter uniformly into the hot side of the element. The efficiency η of the
generator is defined as the ratio of electric power output and heat supplied
per unit time.
The electric power output is the product of the electric current i and
the electric field E, the latter being given by (3.16) with the term ∇(µ
e
/z
e
)
neglected:
P
el
= iE = i
_
l
0
ε(T)(dT/dy)dy −i
2
_
l
0
r(T) dy. (3.21)
The first term is the Seebeck one and the second one describes Joule’s dissipa-
tion. The sign minus in the second term of the right-hand side arises because
in the present example ∇T and i have opposite directions. The total heat flux
˙
Q supplied per unit time at the hot side is the sum of the purely conductive
contribution given by Fourier’s law and the Peltier term πi from (3.15) when
the term µ
e
/z
e
is again neglected. Making use of the second Kelvin relation
(3.17), one may write
3.2 Thermoelectric Effects 77
˙
Q = λ
h
(T
h
−T
c
)l
−1
+iT
h
ε
h
, (3.22)
where subscript “h” means that the mentioned quantities are evaluated at
T = T
h
, in addition, both
˙
Q and i are supposed to be directed downwards.
The efficiency of the thermoelectric generator is given by
η ≡
P
el
˙
Q

iε(T
h
−T
c
) −i
2
rl
λ(T
h
−T
c
)l
−1
+iT
h
ε
. (3.23)
If the difference of temperatures is not very important, in such a way that
the generator may be considered as almost homogeneous, the quantities λ, ε,
and r may be taken as constants, as it has been done in (3.23).
It is usual to express the efficiency η as the product of Carnot’s efficiency
η
Carnot
≡ (T
h
−T
c
)/T
h
and a so-called reduced efficiency η
r
, i.e. η = η
Carnot
η
r
.
We therefore write (3.23) as
η =
T
h
−T
c
T
h
iε −
ri
2
l
T
h
−T
c
λ
T
h
−T
c
lT
h
+iε
. (3.24)
The reduced efficiency may still be expressed in terms of the ratio of fluxes
x = i/[λ(T
h
−T
c
)l
−1
] as
η
r
(x) =
εx −λrx
2
T
−1
h
+εx
=
εx[1 −(xε/Z)]
T
−1
h
+εx
, (3.25)
where Z is the so-called thermoelectric figure of merit
Z =
ε
2

, (3.26)
which has the dimension of the reciprocal of temperature, and depends only
on the transport coefficients, generally function of the temperature. Expres-
sion (3.26) of the figure of merit reflects the property that efficiency is
enhanced by high values of the Seebeck coefficient ε, low values of the
electric resistivity r, and heat conductivity λ. High values of ε contribute
to a strong coupling between heat and electric current and low values
of r and λ minimize the losses due to Joule’s dissipation and to heat
conduction.
In the numerator of (3.25) are competing the thermoelectric effect with
Joule dissipation. Since the first one is linear in the current, or x, and the
second one is non-linear in the same quantity, there exists necessarily an
optimal value of the reduced efficiency. The optimum ratio of the fluxes x
opt
corresponding to the maximum value of η
r
is obtained from the condition

r
(x)/dx = 0; the result is
x
opt
=
(1 +ZT)
1/2
−1
εT
. (3.27)
78 3 Coupled Transport Phenomena
The maximum efficiency is then
η
max
=
T
h
−T
c
T
h
2 +ZT −2

1 +ZT
ZT
. (3.28)
If the ratio x is significantly different from the optimum value (3.27), the
material is not efficiently converting heat energy into electric energy. Once
x is optimized, the optimal current may be found, for a given value of the
temperature gradient. Note that for ZT ¸ 1, one finds back Carnot’s but
materials currently used in thermoelectric devices have relatively low values
of ZT, between 0.4 and 1.5, so that in reality, the maximum Carnot value is
far from being attained.
In general, the temperature difference between the two sides of the genera-
tor will be so high that the assumption of homogeneity used in (3.25)–(3.28)
is not tenable. In this case, one may consider the thermoelectric device as
formed by a series of small quasi-homogeneous elements, at different average
temperatures. For instance, for two elements thermally in series, the com-
bined efficiency is (see Problem 3.9),
η =
P
el,1
+P
el,2
˙
Q
1
= 1 −(1 −η
1
)(1 −η
2
). (3.29a)
In the continuum limit, in which the generator is constituted of many layers
at different temperatures, (3.29a) must be replaced by (see Problem 3.9),
η = 1 −exp
_

_
T
h
T
c
η
r
(x, T)
T
dT
_
. (3.29b)
Using this expression, one finds for the total efficiency of the device
η = 1 −
ε
c
T
c
+x
−1
c
ε
h
T
h
+x
−1
h
. (3.30)
In this configuration, it may happen that the optimum current determined
by x
opt
in one segment (for instance, the hot side) is significantly different
from the optimum value x
opt
in another segment (for instance, the cold side);
in this case, there will be no suitable current for which both parts of the gen-
erator are operating with optimal efficiency. This is a challenge in materials
sciences, as x
opt
is temperature dependent through the transport coefficients
of the material (namely Z and ε) and it would be highly desirable to find ma-
terials with suitable temperature dependence of these coefficients to optimize
the generation.
In terms of the coefficient Z and assuming constant transport coefficients
ε, λ, and r, the maximum efficiency of the power generation is given by (see
Problem 3.7)
η
max
=
T
h
−T
c
T
h
(1 +ZT
av
)
1/2
−1
(1 +ZT
av
)
1/2
+ (T
c
/T
h
)
, (3.31)
3.3 Thermodiffusion: Coupling of Heat and Mass Transport 79
where T
av
is the average temperature T
av
=
1
2
(T
h
+T
c
). When T
c
≈ T
h
, the
ratio T
c
/T
h
in the denominator of (3.31) is close to 1 and the efficiency (3.28)
is recovered.
Analogously, the coefficient of performance for refrigeration (i.e. the ratio
between the heat extracted per unit time and the electric power consumed
by the corresponding engine) is
η =
˙
Q
P
el
=
T
c
T
h
−T
c
(1 +ZT
av
)
1/2
−(T
h
/T
c
)
1 + (1 +ZT
av
)
1/2
. (3.32)
In many practical situations, two parallel generators are used, one of
n-type semiconductors (current carried by electrons, with ε < 0) and an-
other with p-type semiconductors (current brought by holes, with ε > 0).
The global efficiency may be derived from (3.23) or (3.31) by using an aver-
age for both generators, namely
η
np
=
η
p
Q
p

n
Q
n
Q
p
+Q
n
, (3.33)
where Q
p
and Q
n
are the amounts of heat supplied to the hot side of the p
and n elements per unit time.
3.3 Thermodiffusion: Coupling of Heat and Mass
Transport
By thermodiffusion is meant the coupling between heat and matter transport
in binary or multi-component mixtures. In the case of a binary mixture, a
natural choice of the state variables is ρ
1
, ρ
2
(individual mass densities), v
(barycentric velocity), and u (internal energy), but a more convenient choice
is ρ(= ρ
1
+ ρ
2
), c
1
= (ρ
1
/ρ), v, and u. In the classical theory of irreversible
processes, one is more interested by the behaviour of the barycentric velocity
than by the individual velocities of the components and this is the reason why
only the velocity v, and not the individual velocities v
1
and v
2
, figures among
the space of variables. In absence of chemical reactions, viscosity, external
body forces and energy sources, the corresponding evolution equations are
given by

dt
= −ρ∇ v, (3.34)
ρ
dc
1
dt
= −∇ J
1
, (3.35)
ρ
dv
dt
= −∇p, (3.36)
ρ
du
dt
= −∇ q −p∇ v. (3.37)
80 3 Coupled Transport Phenomena
Moreover, we assume that the system is in mechanical equilibrium with zero
barycentric velocity and zero acceleration, which is the case for mixtures
confined in closed vessels. It follows then from (3.34) and (3.36) that the
total mass density and the pressure remain uniform throughout the system
and that the last term in (3.37) vanishes. Substituting the balance equations
of mass fraction (3.35) and internal energy (3.37) in the Gibbs’ equation
ds
dt
= T
−1
du
dt
−T
−1
(¯ µ
1
− ¯ µ
2
)
dc
1
dt
, (3.38)
where use has been made of dc
2
= −dc
1
, results in the following balance of
entropy
ρ
ds
dt
= −∇¦T
−1
[q −(¯ µ
1
−¯ µ
2
)J
1
]¦+q ∇T
−1
−J
1
∇[T
−1
(¯ µ
1
−¯ µ
2
)], (3.39)
with the entropy flux given by
J
s
= T
−1
[q −(¯ µ
1
− ¯ µ
2
)J
1
], (3.40)
and the rate of entropy production by
σ
s
= q ∇T
−1
−J
1
∇[T
−1
(¯ µ
1
− ¯ µ
2
)]. (3.41)
With the help of the Gibbs–Duhem’s relation
c
1
(∇¯ µ
1
)
T,p
+c
2
(∇¯ µ
2
)
T,p
= 0, (3.42)
where subscripts T and p indicate that differentiation is taken at constant T
and p, and the classical result of equilibrium thermodynamics
T∇(T
−1
¯ µ
k
) = −h
k
T
−1
(∇T) + (∇¯ µ
k
)
T
, (3.43)
with h
k
the partial specific enthalpy of substance k(k = 1, 2), we are able to
eliminate the chemical potential from (3.41) of σ
s
, which finally reads as
σ
s
= −q

∇T
T
2

µ
11
Tc
2
J
1
∇c
1
. (3.44)
The new heat flux q

is defined by q

= q −(h
1
−h
2
)J
1
, and it is equal to the
difference between the total flux of heat and the transfer of heat due to diffu-
sion while the quantity µ
11
stands for µ
11
= (∂¯ µ
1
/∂c
1
)
T,p
. The derivation of
(3.44) exhibits clearly the property that the entropy production is a bilinear
expression in the thermodynamic fluxes q

and J
1
and forces taking the form
of gradients of intensive variables, easily accessible to direct measurements.
It is obvious that when the mixture reaches thermodynamic equilibrium, the
heat and mass flows as well as the temperature and mass fraction gradients
3.3 Thermodiffusion: Coupling of Heat and Mass Transport 81
vanish. Expression (3.44) of σ
s
suggests writing the following phenomenolog-
ical relations between fluxes and forces:
q

= −L
qq
∇T
T
2
−L
q1
µ
11
Tc
2
∇c
1
, (3.45)
J
1
= −L
1q
∇T
T
2
−L
11
µ
11
Tc
2
∇c
1
, (3.46)
with the Onsager’s reciprocal relation L
q1
= L
1q
and the following inequali-
ties resulting from the positiveness of entropy production:
L
qq
> 0, L
11
> 0, L
qq
L
11
−L
q1
L
1q
> 0. (3.47)
After introducing the following identifications:
L
qq
T
2
= λ (heat conductivity),
L
11
µ
11
ρc
2
T
= D (diffusion coefficient),
L
q1
ρc
1
c
2
T
2
= D
F
(Dufour coefficient),
L
1q
ρc
1
c
2
T
2
= D
T
(thermal diffusion coefficient),
the phenomenological laws take the form
q

= −λ∇T −ρTµ
11
c
1
D
F
∇c
1
, (3.48)
J
1
= −ρc
1
c
2
D
T
∇T −ρD∇c
1
. (3.49)
Inequalities (3.47) imply in particular that λ > 0 and D > 0 while from
Onsager’s relation is inferred that
D
F
= D
T
. (3.50)
This last result is a confirmation of an earlier result established by Stefan
at the end of the nineteenth century. Starting from the law of conservation
of momentum, Stefan was indeed able to demonstrate the above equality at
least in the case of binary mixtures. By setting the gradient of the mass
fraction equal to zero, (3.48) is identical to Fourier’s equation so that λ can
be identified with the heat conductivity coefficient. Similarly, relation (3.49)
reduces to Fick’s law of diffusion when temperature is uniform and there-
fore D represents the diffusion coefficient; in general, the phenomenological
coefficients in (3.48) and (3.49) are not constant.
In multi-component mixtures, the mass flux of substance i is a linear
function of not only ∇c
i
but also of all the other mass fractions gradients
∇c
j
(j ,= i). Such “diffusion drag” forces have been invoked to interpret some
biophysical effects and play a role in the processes of separation of isotopes
(see Box 3.3).
82 3 Coupled Transport Phenomena
Box 3.3 The Soret’s Effect and Isotope Separation
Thermal diffusion is exploited to separate materials of different molecular
mass. If a fluid system is composed of two kinds of molecules of different
molecular weight, and if it is submitted to a temperature gradient, the
lighter molecules will accumulate near the hot wall and the heavier ones near
the cold wall. This property was used in the 1940s for the separation of
235
U
and
238
U isotopes in solutions of uranium hexafluoride, in the Manhattan
project, leading to the first atomic bomb. Usually this process is carried
out in tall and narrow vertical columns, where convection effects reinforce
the separation induced by thermal diffusion: the light molecules near the
hot walls have an ascending motion, whereas the heavy molecules near
the cold wall sunk towards the lowest regions. This process accumulates
the lightest isotope in the highest regions, from where it may be extracted.
This process is simple but its consumption of energy is high, and therefore
it has been substituted by other methods. However, it is still being used in
heavy water enrichment, or in other processes of separation of light atoms
with the purpose of, for instance, to generate carbide layers on steel, alloys
or cements, thus hardening the surface and making it more resistant to wear
and corrosion. Soret’s effect plays also a role in the structure of flames and in
polymer characterization. More recently, high temperature gradients have
been produced by means of laser beams rather than by heating uniformly
the walls of the container.
The Dufour’s effect, the reciprocal of thermal diffusion, has not so many
industrial applications. It plays nevertheless a non-negligible role in some
natural processes as heat transport in the high atmosphere and in the soil
under isothermal conditions but under a gradient of moisture.
The coefficient D
T
in (3.49) is typical of thermal diffusion, i.e. the flow of
matter caused by a temperature difference; such an effect is referred to as the
Soret’s effect in liquids with the quotient D
T
/D called the Soret coefficient.
The reciprocal effect, i.e. the flow of heat caused by a gradient of concen-
tration as evidenced by (3.48) is the Dufour’s effect. It should be observed
that the cross-coefficients D
T
and D
F
are much smaller than the direct co-
efficients like the heat conductivity λ and the diffusion coefficient D. The
latter turns out to be of the order of 10
−8
m
2
s
−1
in liquids and 10
−5
m
2
s
−1
in gases while the coefficient of thermal diffusion D
T
varies between 10
−12
and 10
−14
m
2
s
−1
K
−1
in liquids and from 10
−8
to 10
−12
m
2
s
−1
K
−1
in gases.
The Soret’s effect is mainly observed in oceanography while the Dufour’s
effect, which is negligible in liquids, has been detected in the high atmosphere.
The smallness of the coupling effects is the reason why they are hard to be
observed and measured with accurateness.
Defining a stationary state by the absence of matter flow (J
1
= 0), it turns
out from (3.49) that
(∆T)
st
= −
D
D
T
c
1
c
2
(∆c
1
)
st
, (3.51)
3.4 Diffusion Through a Membrane 83
which indicates that a difference of concentration is able to generate a tem-
perature difference, called the osmotic temperature. This is typically an irre-
versible effect because the corresponding entropy production is non-zero as
directly seen from (3.41). This effect should not be confused with the os-
motic pressure, which expresses that a difference of concentration between
two reservoirs kept at the same uniform temperature but separated by a
semi-permeable membrane gives raise to a pressure drop, called the osmotic
pressure. The latter is a pure equilibrium effect resulting from the property
that, at equilibrium, the chemical potential ¯ µ(T, p, c
1
) takes the same value
in both reservoirs so that (∆¯ µ)
T
= ∆p/ρ
1

11
∆c
1
= 0, and
∆p = −ρ
1
µ
11
∆c
1
, (3.52)
with ρ
1
the specific mass of the species crossing the membrane; it is directly
checked that in the present situation, the entropy production (3.41) is indeed
equal to zero.
The phenomena studied in this section are readily generalized to multi-
component electrically charged systems, like electrolytes.
3.4 Diffusion Through a Membrane
The importance of transport of matter through membranes in the life of cells
and tissues justify that we spend some time to discuss the problem. In biolog-
ical membranes, one distinguishes generally between two modes of transport:
purely passive transport due to a pressure gradient or a mass concentration
gradient and active transport involving ionic species, electrical currents, and
chemical reactions. Here we focus on some aspects of passive transport. Our
objective is to present a simplified analysis by using a minimum number of
notions and parameters; in that respect, thermal effects will be ignored but
even so, the subject keeps an undeniable utility.
We consider the simple arrangement formed by two compartments I and
II separated by a homogeneous membrane of thickness ∆l, say of the order
of 100 µm. Each compartment is filled with a binary solution consisting of a
solvent 1 and a solute 2 (see Fig. 3.3).
The membrane is assumed to divide the system in two discontinuous sub-
systems that are considered as homogeneous.
3.4.1 Entropy Production
In absence of thermal gradients, it is inferred from (3.41) that the rate of
dissipation, measured per unit volume of the membrane, will take the form
84 3 Coupled Transport Phenomena
Fig. 3.3 System under study consisting in a membrane separating a binary solution

s
= −J
1
∇¯ µ
1
−J
2
∇¯ µ
2
. (3.53)
After integration over the thickness ∆l, the rate of dissipation per unit sur-
face, denoted as Φ, can be written as
Φ = −J
1
∆¯ µ
1
−J
2
∆¯ µ
2
, (3.54)
where ∆¯ µ
i
designates the difference of chemical potential of species i across
the membrane, J
1
and J
2
are the flows of solvent and solute, respectively.
Instead of working with J
1
and J
2
, it is more convenient to introduce the
total volume flow J
V
across the membrane and the relative velocity J
D
of the
solute with respect to the solvent, defined, respectively, by
J
V
= ¯ v
1
J
1
+ ¯ v
2
J
2
, (3.55)
J
D
= v
2
−v
1
, (3.56)
the quantities ¯ v
1
and ¯ v
2
stand for the partial specific volumes of the solvent
and the solute, v
1
and v
2
are their respective velocities given by v
1
= ¯ v
1
J
1
and
v
2
= ¯ v
2
J
2
. With the above choice of variables, (3.54) reads as (Katchalsky
and Curran 1965; Caplan and Essig 1983)
Φ = −J
V
∆p −J
D
∆π, (3.57)
where ∆p = p
I
−p
II
and ∆π = c
2
(∆¯ µ
2
)
p
is the osmotic pressure; the quantity
(∆¯ µ
2
)
p
is that part of the chemical potential depending only on the concen-
tration and defined from ∆¯ µ
2
= V
2
∆p + (∆¯ µ
2
)
p
, c
2
is the number of moles
of the solute per unit volume. For ideal solutions, one has (∆¯ µ
2
)
p
= RT∆c
2
.
3.4 Diffusion Through a Membrane 85
3.4.2 Phenomenological Relations
By assuming linear relations between thermodynamic fluxes and forces, one
has
J
V
= L
VV
∆p +L
VD
∆π, (3.58)
J
D
= L
DV
∆p +L
DD
∆π. (3.59)
The advantage of (3.58) and (3.59) is that they are given in terms of parame-
ters that are directly accessible to experiments. The corresponding Onsager’s
relation is L
VD
= L
DV
, whose main merit is to reduce the number of para-
meters from four to three.
To better apprehend the physical meaning of the phenomenological coeffi-
cients L
VV
, L
DV
, L
DD
, and L
VD
, let us examine some particular experimental
situations. First consider the case wherein the concentration of the solute is
the same on both sides of the membrane such that ∆π = 0. If a pressure
difference ∆p is applied, one will observe according to (3.58) a volume flow
proportional to ∆p; the proportionality coefficient L
VV
is called the mechan-
ical filtration coefficient of the membrane: it is defined as the volume flow
produced by a unit pressure difference between the two faces of the mem-
brane. A further look on relation (3.59) indicates that even in absence of a
concentration difference (∆π = 0), there will be a diffusion flow J
D
= L
DV
∆p
caused by the pressure difference ∆p. This phenomenon is known in colloid
chemistry under the name of ultrafiltration and the coefficient L
DV
is the
ultrafiltration coefficient. An alternative possibility is to impose ∆p = 0 but
different solute concentrations in compartments I and II. In virtue of (3.59),
the osmotic difference ∆π will produce a flow of diffusion J
D
= L
DD
∆π and
L
DD
is identified as the permeability coefficient: it is the diffusional mobility
induced by a unit osmotic pressure ∆π = 1. Another effect related to (3.58)
is the occurrence of a volume flow J
V
= L
VD
∆π caused by a difference of os-
motic pressure at uniform hydrostatic pressure: the coupling coefficient L
VD
is referred to as the coefficient of osmotic flow.
The above discussion has clearly shown the importance of the coupling
coefficient L
DV
= L
VD
; by ignoring it one should miss significant features
about motions across membranes. The importance of this coefficient is still
displayed by the osmotic pressure experiment illustrated by Fig. 3.4.
The two phases I (solvent +solute) and II (solvent alone) are separated by
a semi-permeable membrane only permeable to the solvent. The height of the
solution in the capillary tube gives a measure of the final pressure difference
obtained when the volume flow J
V
vanishes, indeed from (3.58) one obtains
(∆p)
J
V
=0
= −
L
VD
L
VV
∆π. (3.60)
This result indicates that, contrary to what is sometimes claimed, ∆p is
not a measure of the osmotic pressure, this is only true if L
VD
= −L
VV
.
86 3 Coupled Transport Phenomena
Fig. 3.4 Osmotic pressure experiment
This condition is met by so-called ideal semi-permeable membranes whose
property is to forbid the transport of solute whatever the values of ∆p and
∆π. For membranes, which are permeable to the solute, it is experimentally
found that
r ≡ −
L
VD
L
VV
< 1. (3.61)
This ratio that is called the reflection coefficient tends to zero for selec-
tive membranes, like porous gas filters, and has been proposed to act as a
measure of the selectivity of the membrane. For r = 1 (ideal membranes), all
the solute is reflected by the membrane, for r < 1, some quantity of solute
crosses the membrane, while for r = 0 the membrane is completely permeable
to the solute. To clarify the notion of membrane selectivity, let us go back to
the situation described by ∆π = 0. In virtue of (3.58) and (3.59), one has
r = −
L
VD
L
VV
= −
_
J
D
J
V
_
∆π=0
, (3.62)
or, in terms of the velocities introduced in relation (3.56),
_
2v
2
v
1
+v
2
_
∆π=0
= 1 −r. (3.63)
For an ideal semi-permeable membrane (r = 1), one has v
2
= 0 and the
solute will not cross the membrane; for r = 0 (v
1
= v
2
) the membrane is
not selective and allows the passage of both the solute and the solvent; for
negative values of r(v
2
> v
1
), the velocity of the solute is greater than that
of the solvent and this is known as negative anomalous osmosis, which is
a characteristic of the transport of electrolytes across charged membranes.
3.5 Problems 87
Table 3.1 Phenomenological coefficients for two biological membranes
ω L
VV
Membrane Solute Solvent (10
−15
mol dyn
−1
s
−1
) r (10
−11
cm
3
dyn
−1
s
−1
)
Human Methanol Water 122 – –
blood cell
Urea Water 17 0.62 0.92
Toad skin Acetamide Water 4 ×10
−3
0.89 0.4
Thiourea Water 5.7 ×10
−4
0.98 1.1
A final parameter of interest, both in synthetic and biological membranes, is
the solute permeability coefficient
ω =
c
2
L
VV
(L
VV
L
DD
−L
2
VD
). (3.64)
For ideal semi-permeable membranes for which L
VD
= −L
VV
= −L
DD
,
one has ω = 0, and for non-selective membranes (r = 0), it is found that
ω = c
2
L
DD
.
The interest of irreversible thermodynamics is to show clearly that three
parameters are sufficient to describe transport of matter across membrane
and to provide the relationships between the various coefficients character-
izing a semi-permeable membrane. In Table 3.1 are reported some values
of these coefficients for two different biological membranes (Katchalsky and
Curran 1965).
3.5 Problems
3.1. Entropy flux and entropy production. Determine (3.11) and (3.12) of the
entropy flux and the entropy production in the problem of thermoelectricity.
3.2. Onsager’s reciprocal relations. In presence of a magnetic field, Onsager’s
relations can be written as L(H) = L
T
(−H). Decomposing L in a symmetric
and an antisymmetric part L = L
s
+ L
a
, show that L
s
(H) = L
s
(−H) and
L
a
(H) = −L
a
(−H).
3.3. Thermoelectric effects. The Peltier coefficient of a couple Cu–Ni is
π
Cu–Ni
≈ −5.08 mV at 273 K, π
Cu–Ni
≈ −6.05 mV at 295 K, and π
Cu–Ni

−9.10 mV at 373 K. Evaluate (a) the heat exchanged in the junction by
Peltier’s effect when an electric current of 10
−2
A flows from Cu to Ni at
295 K; (b) idem when the current flows from Ni to Cu. (c) The two junctions
of a thermocouple made of Cu and Ni are kept at 305 and 285 K, respec-
tively; by using the Thomson relation, determine the Seebeck coefficient and
estimate the electromotive force developed by the thermocouple.
88 3 Coupled Transport Phenomena
3.4. Thermocouple. The electromotive force ∆ξ of a Cu–Fe thermocouple is
given by ∆ξ(µV ) ≈ −13.4∆T − 0.01375(∆T)
2
, where ∆T = T
R
− T, the
reference temperature being T
R
= 273 K. (a) Evaluate the Peltier heat ex-
changed in a Cu–Fe junction at 298 K when an electric current of 10
−3
A flows
from Cu to Fe. (b) By using the Thomson relation, evaluate the difference of
Thomson coefficients of these two metals at 298 K.
3.5. Evaluation of the maximum efficiency of a thermoelectric element.
Assume a finite temperature difference and constant transport coefficients.
(a) Starting from the definition of the ratio of fluxes x ≡ i/λ∇T, and from
the steady state form of (3.18), prove that
dx
−1
dT
= T

dT
−xλr.
(b) Taking into account this relation, and for ε independent of T, show that
the values of x at the cold and hot boundaries of the systems, x
c
and x
h
,
respectively, may be written in terms of x
opt,av
(Snyder and Ursell 2003)
x
−1
h
= x
−1
opt,av

1
2
∆T(λr)x
opt,av
,
x
−1
c
= x
−1
opt,av
+
1
2
∆T(λr)x
opt,av
,
where x
opt,av
is the value for x
opt
found in (3.27) for the average temperature
T
av
=
1
2
(T
h
+T
c
). (c) Introducing these values in (3.30), show finally (3.31),
i.e.
η =
T
h
−T
c
T
h
(1 +ZT
av
)
1/2
−1
(1 +ZT
av
)
1/2
+ (T
c
/T
h
)
.
3.6. The figure of merit of a thermoelectric material. The figure of merit
for thermoelectric materials is defined by Z ≡ ε
2
/λr, where ε is the Seebeck
coefficient, r is the electrical resistivity, and λ is the thermal conductivity. (a)
Check that the dimension of this combination is the inverse of temperature.
(b) The figure of merit of a junction of two thermoelectric materials in a
thermocouple is defined as
Z
tc

ε
2
AB
[(λ/σ)
1/2
A
+ (λ/σ)
1/2
B
]
2
.
Show that the efficiency of the thermocouple, defined as the ratio of the
electric power delivered and the rate of heat supplied to the hot junction is
η
tc
=
T
1
−T
2
3T
1
+T
2
2
+
4
Z
tc
.
3.7. Heat engines in series. Consider two Carnot heat engines in series: the
first one works between heat reservoirs at T
1
and T
2
and the second one
between reservoirs at T
2
and T
3
, assume in addition that the whole amount
3.5 Problems 89
of heat Q
2
delivered by the first engine to the reservoir T
2
is transferred to
the second engine. (a) Show that one recovers (3.31), namely
η =
P
el,1
+P
el,2
Q
1
= 1 −(1 −η
1
)(1 −η
2
).
(b) In (3.29b) it has been considered the continuum limit of high number of
Carnot engines working between reservoirs in series at temperatures T
i
+dT
and T
i
, with T
i
the temperature of the cold reservoir corresponding to ith
Carnot engine, T
i
ranging from T
c
to T
h
. Show that the previous expression
leads to (3.29b) by writing
η = 1 −

i=1
(1 −η
i
) ≈ 1 −exp
_

_
T
h
T
c
η(T)
T
dT
_
.
3.8. Thermophoresis. Thermophoresis in a quiescent fluid is described by the
phenomenological equation
J = −D[∇c +S
T
c(1 −c)∇T],
where D is the diffusion coefficient and S
T
≡ D
T
/D is the Soret coefficient,
D
T
being the thermal diffusion coefficient. (a) Show that for concentrations
much less than unity, the concentration of the solute at temperature T is
given by
c = c
0
exp[−S
T
(T −T
0
)],
where c
0
is the concentration at temperature T
0
. For a DNA sample, it was
found that S
T
≈ 0.14 K
−1
around T = 297 K (Braun and Libchaber 2002).
(b) Calculate c/c
0
for two regions with a temperature difference ∆T = 2.5 K.
3.9. Cross-effects in membranes. Assume that the flows of water J
1
and solute
J
2
across a membrane are, respectively, expressed as
J
1
= L
11
∆p +L
12
∆c
2
,
J
2
= L
21
∆p +L
22
∆c
2
,
where ∆p and ∆c
2
are, respectively, the differences of pressure and solute
concentration between both sides of the membrane. Is the Onsager’s recipro-
cal relation L
12
= L
21
applicable? Explain.
3.10. Transport of charged ions across membranes. The Nernst equation.
Transport of charged ions, mainly H
+
, Na
+
, K
+
, Ca
2+
, and Cl

, across mem-
branes plays a crucial role in many biological processes. Consider a membrane
with both sides at concentrations c
in
, c
out
, and voltages 1
in
, 1
out
. An argu-
ment similar to that leading to (3.58) and (3.59) yields, in the isothermal
case, the following result for the diffusion flux J (number of ions which cross
the membrane per unit area and unit time)
90 3 Coupled Transport Phenomena
J = −
˜
D(∇µ +q∇1),
q designating the electrical charge. (a) Show that the voltage difference for
which both sides of the membrane are at equilibrium (J = 0) with respect
to the transport of a given ionic species i (assumed to behave as an ideal
substance, i.e. µ = µ
(0)
+k
B
T ln c) is given by
(1
in
−1
out
)
eq
=
k
B
T
q
ln
c
out
c
in
.
This equation is known as Nernst equation. (b) Assume that the membrane is
immersed in solutions of NaCl–water with concentration 1 M (where M stands
for mole per litre) and 0.1 M and that only Na
+
may cross the membrane.
Calculate the difference of voltage between both sides of the membrane at
equilibrium. (c) In typical biological cells at rest, the K
+
concentration are
c
in
= 150 mM, c
out
= 5 mM, and the voltages are V
out
= 0 mV, V
in
= −70 mV.
Evaluate the Nernst potential for K
+
at T = 310 K, and determine whether
K
+
will move towards or outwards the cell.
3.11. Transport of charged ions across membranes: the Goldmann equation.
In the derivation of the Nernst equation, it is assumed that the membrane
is only permeable to one ionic species. In actual situations, several ionic
species may permeate through the membrane. If one considers the transport
of several ionic species as K
+
, Na
+
, and Cl

, which are the most usual ones,
the equilibrium voltage difference is given by the Goldmann equation
(1
in
−1
out
)
eq
=
k
B
T
q
ln
P
K
c
K,out
+P
Na
c
Na,out
+P
Cl
c
Cl,in
P
K
c
K,in
+P
Na
c
Na,in
+P
Cl
c
Cl,out
,
where P
K
, P
Na
, and P
Cl
refer to the relative permeabilities of the correspond-
ing ions across the membrane. Note that when only one species crosses the
membrane (i.e. when the permeabilities vanish for two of the ionic species),
the Goldmann equation reduces to the Nernst relation. (a) Derive the Gold-
mann equation. Hint: Take into account the electroneutrality condition. (b)
It is asked why the Cl concentrations appear in the equation in a different
way as the concentrations of the other two ions, namely c
in
instead of c
out
and vice versa. (c) In the axon at rest, i.e. in the long cylindrical terminal
of the neurons along which the output electrical signals may propagate, the
permeabilities are P
K
≈ 25P
Na
≈ 2P
Cl
, whereas at the peak of the action
potential (the electric nervous signal), P
K
≈ 0.05P
Na
≈ 2P
Cl
. The dramatic
increase in the sodium permeability is due to the opening of sodium chan-
nels when the voltage difference is less than some critical value. Evaluate
the Goldmann potential in both situations, by taking for the concentrations:
c
K,in
= 400 mM, c
K,out
= 20 mM, c
Na,in
= 50 mM, c
Na,out
= 440 mM, and
c
Cl,in
= 50 mM, c
Cl,out
= 550 mM.
Chapter 4
Chemical Reactions and Molecular
Machines
Efficiency of Free-Energy Transfer and Biology
Chemical reactions are among the most widespread processes influencing life
and economy. They are extremely important in biology, geology, environmen-
tal sciences, and industrial developments (energy management, production
of millions of different chemical species, search for new materials). Chemical
kinetics is a very rich but complex topic, which cannot be fully grasped by
classical irreversible thermodynamics, whose kinetic description is restricted
to the linear regime, not far from equilibrium.
Despite this limitation, this formalism is able to yield useful results, es-
pecially when dealing with coupled reactions. Indeed, the study of coupled
processes is one of the most interesting features of non-equilibrium thermo-
dynamics. In particular, the various couplings – between several chemical
reactions, between chemical reactions and diffusion, and between chemical
reactions and active transport processes in biological cells – receive a simple
and unified description. Non-equilibrium thermodynamics is able to provide
clarifying insights to some conceptually relevant problems, particularly in
biology, which are rarely investigated in other formalisms.
In that respect, in the forthcoming we will focus on the problem of ef-
ficiency of free-energy transfer between coupled reactions, and the descrip-
tion of biological molecular motors. These engines may be modelled by some
particular cycles of chemical reactions, generalizing the triangular chemical
scheme proposed by Onsager in his original derivation of the reciprocal re-
lations. Special attention will also be devoted to the effects arising from the
combination of chemical reactions and diffusion. From a biological perspec-
tive, this coupling provides the basis of cell differentiation in the course of
development of living organisms. The ingredients of this extremely complex
phenomenon, leading to spatial self-organization far from equilibrium, may
be qualitatively understood by considering a simplified version of the mech-
anism of coupling between autocatalytic chemical reactions and diffusion.
Biological applications of non-equilibrium thermodynamics have been the
subject of several books (Katchalsky and Curran 1965; Nicolis and Prigogine
1977; Hill 1977; Caplan and Essig 1983; Westerhoff and van Dam 1987; Jou
91
92 4 Chemical Reactions and Molecular Machines
and Llebot 1990; Nelson 2004). Here, we present a synthesis of the main ideas
and, for pedagogical purpose, we start the analysis with the problem of one
single chemical reaction.
4.1 One Single Chemical Reaction
Equilibrium thermodynamics provides a general framework to analyse the
equilibrium conditions of chemical reactions, and it leads in a natural way
to the concept of equilibrium constant and its modifications under changes
of temperature and pressure, as recalled in Sect. 1.7. A clear understanding of
these effects is crucial to establish the range of temperature and pressure for
the optimization of industrial processing. However, this optimization cannot
be carried out without taking into account kinetic effects. For instance, low-
ering temperature has generally as a consequence to slow down the reaction
velocity. On the other side, it is known from equilibrium thermodynamics
that, for exothermic reactions, the efficiency of conversion of reactants into
products increases when temperature is lowered. Thus, lowering tempera-
ture has two opposite effects: a lowering of the velocity of reactions and an
increase of efficiency. Optimization will therefore result from a compromise
between equilibrium factors and kinetic factors. We will return to this topic
in Chap. 5. Here, we emphasize some kinetic aspects of chemical reactions,
from the point of view of non-equilibrium thermodynamics.
To illustrate our approach consider, as in Chap. 1, the reaction of synthesis
of hydrogen chloride
H
2
+ Cl
2
2HCl. (4.1)
Since we will work in terms of local quantities, it is convenient to introduce
the mass fractions c
k
defined as c
k
≡ m
k
/m, with m
k
the mass of component
k and m the total mass. In virtue of the law of definite proportions (1.70),
the change of c
k
during the time interval dt may be written as
ρ
dc
k
dt
= ν
k
M
k

dt
, k = 1, 2, . . ., n, (4.2)
with ν
k
the stoichiometric coefficient of species k, M
k
its molar mass, and
dξ/dt the velocity of reaction per unit volume. This equation can be deduced
from (1.70) when the mole numbers are expressed in terms of mass fractions
and the degree of advancement is given per unit volume.
Besides the mass of the n species, the other relevant variable is the specific
internal energy u, because of the exchange of heat with the outside. The time
evolution equation for the internal energy u is given by the first law
ρ
du
dt
= −∇ q, (4.3)
4.1 One Single Chemical Reaction 93
where q is the heat flux vector expressing the amount of heat exchanged
between the system and the outside world acting as a reservoir.
Now, we want to obtain expressions for the velocity of the reaction in
terms of the corresponding thermodynamic force. This will be achieved within
the general framework set out in Chap. 2, demanding that we previously
determined the expression of the entropy production. To determine it, we
start from the Gibbs’ relation
ds
dt
=
1
T
du
dt

k
¯ µ
k
T
dc
k
dt
, (4.4a)
where ¯ µ
k
is the chemical potential of the species k measured per unit mass.
Making use of (4.2) and the relation M
k
¯ µ
k
= µ
k
, with µ
k
the chemical
potential per unit mole, one obtains
ρ
ds
dt
=
1
T
ρ
du
dt

k
ν
k
µ
k
T

dt
. (4.4b)
We know from Sect. 1.7 that for ideal systems, i.e. mixture of ideal gases or
dilute solutions, the chemical potential per unit mole µ
k
is given by
µ
k
= µ
(0)
k
(T, p) +RT ln x
k
, (4.5)
where x
k
≡ N
k
/N is the mole fraction while µ
(0)
k
is independent of the
composition.
At this stage, it is relevant to introduce the notion of affinity defined in
(1.76) as
/ ≡ −

k
ν
k
µ
k
. (4.6)
As seen below, the reaction will proceed forwards if / > 0 or backwards
if / < 0: hence the name of affinity. At fixed values of temperature and
pressure, / is only function of the mole fractions of species, and it was shown
in Sect. 1.7 that
/ = RT ln
K(T, p)
x
ν
1
1
x
ν
2
2

, (4.7a)
with K(T, p) the equilibrium constant. Note that this constant may also be
written as K(T, p) = (x
ν
1
1
)
eq
(x
ν
2
2
)
eq
, because at equilibrium / = 0 and
then the term inside the logarithm must be 1. It is useful, for further purposes
in this chapter, to rewrite (4.7a) as
/ = RT ln
(x
ν
1
1
)
eq
(x
ν
2
2
)
eq

x
ν
1
1
x
ν
2
2

. (4.7b)
A third way of expressing the affinity is in terms of the molar concentrations,
i.e. in the number of moles per unit volume, namely ˜ c
k
≡ N
k
/V . In this case,
(4.7b) may be expressed as
94 4 Chemical Reactions and Molecular Machines
/ = RT ln
(˜ c
ν
1
1
)
eq
(˜ c
ν
2
2
)
eq

˜ c
ν
1
1
˜ c
ν
2
2

. (4.7c)
To derive the expression of the rate of entropy production, we introduce
the balance equations of mass (4.2) and energy (4.3) in Gibbs’ relation (4.4a)
and (4.4b). Since in chemical reactions, the temperature is generally assumed
to remain uniform, one obtains
ρ
ds
dt
= −∇
q
T
+
/
T

dt
. (4.8)
By comparing this expression with the general form of the entropy balance
equation ρds/dt = −∇ J
s

s
, it follows that the expression of the entropy
flux is
J
s
=
q
T
, (4.9)
while the entropy production is identified as
σ
s
=
/
T
w > 0, (4.10)
i.e. a bilinear expression in the thermodynamic force //T and the flux w =
dξ/dt, which is the velocity of reaction. Truly, the reaction must be considered
as the net effect of a forward flux w
+
going from reactants to products minus
a backward flux w

going from products to reactants, namely w = w
+
−w

.
According to the second law, σ
s
is always a positive quantity and zero
in equilibrium. Thus when / is positive (respectively, negative), the velocity
w = w
+
−w

is also positive (negative) and the reaction will take place from
left to right (right to left). At equilibrium, the affinity / is zero and the net
flux w will be zero, i.e. the flux w
+
towards the right is equal to the flux w

towards the left.
For a single chemical reaction, the phenomenological relation between flux
and force in the linear regime is
w = l
/
T
, (4.11)
where l is a phenomenological coefficient, which in (4.16) is related to the
microscopic rate constants. On the condition that l > 0, this is the simplest
way to guarantee the positiveness of the corresponding entropy production
given by
σ
s
= l
_
/
T
_
2
. (4.12)
However, it should be stressed that the linear flux–force relation (4.11) is
only valid close to equilibrium. This is easily seen by comparing this result
with the classical kinetic expression for the reaction rate which, in the case
of the synthesis of HCl, reads as
w = w
+
−w

= k
+
˜ c
H
2
˜ c
Cl
2
−k

˜ c
2
HCl
= k
+
˜ c
H
2
˜ c
Cl
2
_
1 −
1
K
˜ c
2
HCl
˜ c
H
2
˜ c
Cl
2
_
, (4.13)
4.1 One Single Chemical Reaction 95
where k
+
and k

are, respectively, the rate constants corresponding to the for-
ward and backward reactions, and K = k
+
/k

= (˜ c
2
HCl
)
eq
/[(˜ c
H
2
)
eq
(˜ c
Cl
2
)
eq
].
In terms of the affinity, and in virtue of (4.7c), relation (4.13) can be ex-
pressed as
w = k
+
˜ c
H
2
˜ c
Cl
2
[1 −exp(−//RT)] . (4.14)
For reactions near equilibrium, the affinity / is very small and it is justified
to make a Taylor expansion around / = 0 so that
w = k
+
(˜ c
H
2
)
eq
(˜ c
Cl
2
)
eq
R
/
T
. (4.15)
This result indicates clearly that the linear flux–force law (4.11) is only true
very close to equilibrium and that the classical theory of irreversible processes
does not cope with situations far from equilibrium. By comparison of the
phenomenological relation (4.11) with the kinetic corresponding law (4.15),
we are able to identify the phenomenological coefficient l as
l = k
+
(˜ c
H
2
)
eq
(˜ c
Cl
2
)
eq
R
, (4.16)
which confirms that l is a positive quantity as all the factors of (4.16) are
positive.
The phenomenological relation (4.11) allows us to determine the behaviour
of the degree of advancement ξ in the course of time. Indeed, if / is only
depending on ξ, one can write (4.11) under the form

dt
=
l
T
/ =
l
T
_
/
eq
+
_
∂/
∂ξ
_
(ξ −ξ
eq
)
_
=
l
T
_
∂/
∂ξ
_
(ξ −ξ
eq
), (4.17)
after use is made of /
eq
= 0. This result suggests defining a relaxation time
τ by
τ ≡ −
T
l(∂//∂ξ)
eq
> 0. (4.18)
In this expression, there appear two different factors: one of them is the kinetic
phenomenological coefficient l, whose positiveness is demanded by the second
law; the other one (∂//∂ξ)
T
is related to the thermodynamic stability of the
system and is negative, as shown in Sect. 1.7.
Finally, introducing (4.18) in relation (4.17) leads to
τ

dt
= −(ξ −ξ
eq
) (t > 0), (4.19)
whose solution is an exponentially decreasing function
ξ −ξ
eq
= (ξ
0
−ξ
eq
) exp(−t/τ), (4.20)
where ξ
0
is the initial value of the degree of advancement. Within the limit
of an infinite time, the system tends of course to its equilibrium state.
96 4 Chemical Reactions and Molecular Machines
Equation (4.20) is particularly interesting, because it provides some general
information about the relaxational process without a detailed knowledge of
the kinetics. If the stability condition (∂//∂ξ)
eq
< 0 is not satisfied, the
relaxation time will be negative, indicating that the degree of advancement
will not relax to equilibrium, but will increase exponentially.
4.2 Coupled Chemical Reactions
The study of coupled reactions is not merely a formal generalization of the
above considerations, it is also the basis of discussion of various relevant
concepts. For instance, if two processes are coupled, one of them may proceed
in the opposite direction it would have if it were alone. We will stress this
important feature in the next sections. Another interesting application of the
coupling between chemical reactions is provided by the triangular scheme
A B C A which will serve as a basis to establish Onsager’s reciprocal
relations.
4.2.1 General Formalism
The main modifications with respect to the results of Sect. 4.1 are the expres-
sions of the mass balance and the entropy production. Indeed, the variation
of the mass fractions c
k
of a given species is no longer due to one single
reaction, but depends on the several chemical reactions in which the given
species participates; it is now expressed as
dc
k
dt
=
r

j=1
ν
kj
M
k

j
dt
, k = 1, 2, . . ., n, (4.21)
where subscript j refers to the r different chemical reactions, each of which
has its own degree of advancement ξ
j
, and ν
kj
is the stoichiometric coefficient
of substance k in reaction j. By following the same steps as in Sect. 4.1, it is
found that, under isothermal conditions, the entropy production is
σ
s
=

j
/
j
T
w
j
. (4.22)
It is worth noting that the second law imposes that the whole sum in (4.22),
but not that each individual term, must be positive. Some terms may indeed
be negative, provided that their sum remains positive. Within the assumption
of linear flux–force relations, one has
w
i
=

j
L
ij
/
j
T
, (4.23)
4.2 Coupled Chemical Reactions 97
with, in virtue of Onsager’s reciprocal relations,
L
ij
= L
ji
. (4.24)
In Sect. 4.2.2, a particular derivation of these symmetry relations for a trian-
gular reaction scheme is presented.
4.2.2 Cyclical Chemical Reactions and Onsager’s
Reciprocal Relations
A demonstration of the reciprocal relations (4.24), based on the cycle of
chemical reactions A B C A depicted in Fig. 4.1, was proposed by
Onsager in his celebrated papers of 1931. In Sect. 4.4, we will use a modified
form of this scheme as a model for chemically driven molecular motors.
Let k
i
(i = 1, 2, 3) be the kinetic constants, w
i
the respective velocities of
the reactions, which, according to the mass action law, are given by w
1
=
k
1
˜ c
A
−k
−1
˜ c
B
, w
2
= k
2
˜ c
B
−k
−2
˜ c
C
, w
3
= k
3
˜ c
C
−k
−3
˜ c
A
, and /
i
the respective
affinities, i.e. /
1
= µ
A
− µ
B
, /
2
= µ
B
− µ
C
, /
3
= µ
C
− µ
A
; for simplicity,
all the stoichiometric coefficients are supposed to be equal to one. Since the
process is cyclic, only two reactions are independent and /
1
+/
2
+/
3
= 0.
It will be shown that, when the flux–force relations are of the form
w
1
−w
3
= L
11
/
1
+L
12
/
2
, (4.25a)
w
2
−w
3
= L
21
/
1
+L
22
/
2
, (4.25b)
the Onsager’s reciprocal relation L
12
= L
21
is automatically satisfied near
equilibrium, if the principle of detailed balance is valid. This principle states
Fig. 4.1 Triangular reaction scheme
98 4 Chemical Reactions and Molecular Machines
that transition A to B has the same probability as transition B to A, from
which it follows that, in equilibrium, w
1
= w
2
= w
3
= 0. Note that this
assumption does not follow directly from the phenomenological equations,
since non-zero but equal values w
1
= w
2
= w
3
,= 0 are compatible with
/
1
= /
2
= /
3
= 0. The mathematical details of the proof that L
12
= L
21
are found in Box 4.1.
Box 4.1 Mathematical Details of Onsager’s Demonstration
In virtue of (4.7c), the affinities /
1
and /
2
may be written as
/
1
= RT ln
_
˜ c
A
(˜ c
A
)
eq
(˜ c
B
)
eq
˜ c
B
_
, /
2
= RT ln
_
˜ c
B
(˜ c
B
)
eq
(˜ c
C
)
eq
˜ c
C
_
, (4.1.1)
where use has been made of the results K
1
(T, p) = (˜ c
B
)
eq
/(˜ c
A
)
eq
and
K
2
(T, p) = (˜ c
C
)
eq
/(˜ c
B
)
eq
. By assuming that ˜ c
i
= (˜ c
i
)
eq
+ δ˜ c
i
, with
δ˜ c
i
¸ (˜ c
i
)
eq
, the affinities may be expressed up to the first order in the
deviations δ˜ c
i
from equilibrium as
/
1
= RT
_
δ˜ c
A
(˜ c
A
)
eq

δ˜ c
B
(˜ c
B
)
eq
_
, /
2
= RT
_
δ˜ c
B
(˜ c
B
)
eq

δ˜ c
C
(˜ c
C
)
eq
_
, (4.1.2)
and, from /
3
= −(/
1
+/
2
),
/
3
= RT
_
δ˜ c
C
(˜ c
C
)
eq

δ˜ c
A
(˜ c
A
)
eq
_
. (4.1.3)
Furthermore, according to the mass action law, one has
w
1
−w
3
= (k
1
+k
−3
)˜ c
A
−k
−1
˜ c
B
−k
3
˜ c
C
, (4.1.4)
w
2
−w
3
= k
2
˜ c
B
+k
−3
˜ c
A
−(k
−2
+k
3
)˜ c
C
. (4.1.5)
We now take into consideration the restrictions coming from the detailed
balance. From w
1
= w
2
= w
3
= 0, it is inferred that
k
1
(˜ c
A
)
eq
= k
−1
(˜ c
B
)
eq
, k
2
(˜ c
B
)
eq
= k
−1
(˜ c
C
)
eq
, k
3
(˜ c
C
)
eq
= k
−3
(˜ c
A
)
eq
.
(4.1.6)
Using (4.1.6) to eliminate k
−1
, k
−2
, and k
−3
in (4.1.4) and (4.1.5), one
obtains
w
1
−w
3
= k
1
(˜ c
A
)
eq
_
δ˜ c
A
(˜ c
A
)
eq

δ˜ c
B
(˜ c
B
)
eq
_
+k
3
(˜ c
C
)
eq
_
δ˜ c
A
(˜ c
A
)
eq

δ˜ c
C
(˜ c
C
)
eq
_
,
(4.1.7)
w
2
−w
3
= k
2
(˜ c
B
)
eq
_
δ˜ c
B
(˜ c
B
)
eq

δ˜ c
C
(˜ c
C
)
eq
_
+k
3
(˜ c
C
)
eq
_
δ˜ c
A
(˜ c
A
)
eq

δ˜ c
C
(˜ c
C
)
eq
_
.
(4.1.8)
4.2 Coupled Chemical Reactions 99
In view of the results (4.1.2) and (4.1.3), the above equations may
be written as
w
1
−w
3
=
k
1
RT
(˜ c
A
)
eq
/
1

k
3
RT
(˜ c
C
)
eq
/
3
=
1
RT
[k
1
(˜ c
A
)
eq
+k
3
(˜ c
C
)
eq
]/
1
+
k
3
RT
(˜ c
C
)
eq
/
2
, (4.1.9)
w
2
−w
3
=
k
2
RT
(˜ c
B
)
eq
/
2

k
3
RT
(˜ c
C
)
eq
/
3
=
k
3
RT
(˜ c
C
)
eq
/
1
+
1
RT
[k
2
(˜ c
B
)
eq
+k
3
(˜ c
C
)
eq
]/
2
, (4.1.10)
where use has been made of /
3
= −(/
1
+ /
2
). Note that (4.1.9) and
(4.1.10) have been derived exclusively by reference to chemical kinetics and
the principle of detailed balance.
A look on the coefficients of /
2
in (4.1.9) and /
1
in (4.1.10), which referring
to (4.25a) and (4.25b), can be identified as L
12
and L
21
, respectively, leads
to the conclusion that
L
12
= L
21
. (4.26)
Let us now determine the corresponding results from classical irreversible
thermodynamics. The entropy production σ
s
is given by

s
= w
1
/
1
+w
2
/
2
+w
3
/
3
, (4.27)
or, in terms of independent affinities,

s
= (w
1
−w
3
)/
1
+ (w
2
−w
3
)/
2
, (4.28)
which suggests the following constitutive relations:
w
1
−w
3
= L
11
/
1
+L
12
/
2
, (4.29)
w
2
−w
3
= L
21
/
1
+L
22
/
2
. (4.30)
Comparison with relations (4.1.9) and (4.1.10) yields
L
12
= L
21
=
k
3
RT
(˜ c
C
)
eq
, (4.31)
which completes the proof that the reciprocal relations are equivalent to the
principle of detailed balance.
It should be realized that these results are only valid near equilibrium,
where the net flux along the cycle is zero. In molecular machines, the above
results are not directly transposable because the net flux along the cycle is
different from zero.
100 4 Chemical Reactions and Molecular Machines
4.3 Efficiency of Energy Transfer
A particularly appealing problem in non-equilibrium thermodynamics is to
determine the efficiency with which free energy is exchanged between coupled
chemical reactions. Indeed, in many biological situations, the transfer of free
energy (or free-energy transduction, as is often referred to in textbooks on
biochemistry) is of decisive importance. It is particularly true in microscopic
molecular motors – one of the frontiers of current research in biophysics –
which take free energy from some process (for instance, a chemical reaction)
and convert it into work (for instance, to carry out another chemical reaction
or a transport process). Note, however, that this subject cannot be studied
from the classical perspective of thermal engines, because biological systems
are often at homogeneous temperature, so that Carnot’s efficiency would be
zero. In contrast, the essential ideas of such isothermal free-energy transfer
can be well interpreted in the framework of linear non-equilibrium thermo-
dynamics.
Consider two coupled reactions, for which the dissipated energy is

s
= w
1
/
1
+w
2
/
2
. (4.32)
The corresponding phenomenological equations describing the kinetics of the
reactions are
w
1
= L

11
/
1
+L

12
/
2
, (4.33)
w
2
= L

21
/
1
+L

22
/
2
. (4.34)
The above results remain applicable when one of the processes, say 2, is a
transfer of matter as considered in Sect. 4.4; such a situation is frequent in
biological cells. To guarantee the positiveness of Tσ
s
, either both terms at the
right-hand side of (4.32) are positive or one of them, say w
2
/
2
, is negative
but its absolute value is smaller than the first one, assumed to be positive.
As a consequence, process 2 will evolve in opposite direction of its natural
tendency, for instance mass transport from a region of low to high concentra-
tion. A relevant question is the efficiency of the energy transfer from the first
to the second reaction. Such kind of coupling is frequently met in biology.
For instance, in oxidative phosphorylation, the free energy delivered from the
oxidation of a substrate will serve to the phosphorylation of ADP (adenosine
diphosphate) to give ATP (adenosine triphosphate), which is a basic mole-
cule for metabolic energy exchanges. How much energy is transferred from
oxidation to phosphorylation is an interesting problem or, stated more gen-
erally, what is the efficiency of energetic conversion (Katchalsky and Curran
1965; Jou and Llebot 1990; Westerhoff and van Dam 1987; Criado-Sancho
and Casas-V´azquez 2004).
In the above problem, reaction 1 is supplying some free energy, a part
of which is taken up by reaction 2 to proceed against its usual direction,
whereas the remaining part is lost under the form of heat. The objective
4.3 Efficiency of Energy Transfer 101
is to obtain the maximum efficiency from such an energy transfer (or en-
ergy transduction). The problem is parallel to that of Carnot in the context
of thermal engines. In the latter, heat was partially converted into work
and the remaining was dissipated. Here, the energy delivered by one reac-
tion is partially taken up by the other reaction and the remaining part is
dissipated.
In the following, we suppose that the fluxes w
1
and w
2
are both positive
and that the forces are of opposite signs: /
1
> 0 and /
2
< 0. The degree of
coupling of the two processes is conveniently quantified by the coefficient q
defined by
q ≡
L

12
(L

11
L

22
)
1/2
. (4.35)
From the requirement that the dissipated energy is positive definite, one has
L

11
L

22
≥ (L

12
)
2
, and it follows that q will take values between −1 and +1:
for q = 1, the processes are completely coupled; for q = 0, they are totally
uncoupled. When q < 0, the processes are not compatible in the sense that
process (1) with w
1
/
1
> 0 is unable to produce a process like (2) with a
positive flux w
2
and a negative force /
2
.
In relation with coupled processes, two further quantities have been in-
troduced: the stoichiometric coefficient w
2
/w
1
and the efficiency η of the
conversion defined by
η ≡ −
w
2
/
2
w
1
/
1
. (4.36)
Introducing the quantities x ≡ /
2
//
1
, Z ≡ (L

22
/L

11
)
1/2
, and using (4.33)
and (4.34), it is easily checked that
η = −
Zx(q +Zx)
1 +qZx
, (4.37)
which is maximum, namely, dη/dx = 0, at
x
max
=
(1 −q
2
)
1/2
−1
qZ
, (4.38)
thus leading to
η
max
=
q
2
[1 + (1 −q
2
)
1/2
]
2
. (4.39)
This result is worth to be underlined: in some aspects, it can be compared
with the classical Carnot’s result for the maximum efficiency of heat en-
gines, as it puts out an upper bound for the efficiency of energy transfer
between both reactions or, in general terms, between two coupled processes.
In heat engines, Carnot’s maximum efficiency depends only on the values of
the temperature of the heat reservoirs; here, the result (4.39) has a similar
universality: it depends only on the coupling coefficient but not on the indi-
vidual phenomenological coefficients. It is observed that the maximum of η
102 4 Chemical Reactions and Molecular Machines
−1.0 −0.8 −0.6 −0.4 −0.2 0.0
0.0
0.2
0.4
0.6
0.8
1.0
1.00
0.99
0.95
0.90
0.80
0.70
η
Zx
0.60
Fig. 4.2 Efficiency η vs. the parameter Zx for several values of the coupling coeffi-
cient q
diminishes with decreasing values of q and vanishes for q = 0. This is logical
because, in absence of coupling, the efficiency of the energy transfer should
be zero: this means that all the free energy delivered by reaction 1 would be
dissipated into heat, so that reaction 2 cannot take place against its usual
direction.
In Fig. 4.2, the values of the efficiency η as a function of the quantity Zx
for several values of the coupling coefficient are reported. For q = 1, the
maximum of η is 1; in virtue of (4.36), this situation corresponds to w
1
/
1
=
−w
2
/
2
and consequently to a zero entropy production (reversible process).
The above considerations provide a clearer insight on the nature and im-
portance of the various phenomenological coefficients, and they may be ex-
tended to cover a wider domain of situations, as the thermoelectric generators
in Sect. 3.2 (Van den Broeck 2005).
4.4 Chemical Reactions and Mass Transport:
Molecular Machines
The general developments of Sect. 4.3 will be applied to the study of molecular
machines, mainly pumps across membranes and motors along filaments which
are briefly introduced in Box 4.2. In classical non-equilibriumthermodynamics,
4.4 Chemical Reactions and Mass Transport: Molecular Machines 103
Box 4.2 More About Biological Molecular Engines
An illustration of coupling between chemical reactions and transport is the
transport of macromolecules or vesicles along filaments of microtubules by
kinesin or dinein, two well-known molecular motors; these processes con-
tribute also to the formation of other organized intracellular motions, for
example the separation of chromosomes during cell division (see, for in-
stance, Nelson 2004). These motors must overcome the viscous resistance,
and this is achieved by using the free energy of chemical reactions. Another
example of chemically driven motion is that of the actin/myosin system
in myofibrils, to which is due the muscle motility. In this system, bundles
of myosin molecules are interleaved with actin filaments; upon activation,
myosin crawls along the actin fibres, shortening the muscle fibre, the process
being fuelled by the hydrolysis of ATP.
Most cells are characterized by membrane pumps for the active trans-
port of Na
+
(towards the exterior) and K
+
(towards the interior); this is
especially important in nerve cells, wherein the action potential is due to a
transfer of Na
+
and K
+
across the excited membrane of the axon of neu-
rons. Other kinds of pumps in biological cells are pumps of Ca
2+
which
are found in muscles, and also in the synapses, and pumps of H
+
, which
are important in oxidative phosphorylation in mitochondria and in photo-
synthesis in chloroplasts. On the other side, the flow of H
+
across some
molecular engines may gear molecular motors, as for instance the flagellar
motor of some bacteria, which convert the electrochemical potential jump
of protons into mechanical torque for swimming.
under linear and isotropic conditions, chemical reactions (scalar processes) are
not coupled with mass transport (a vector process). However, the presence
of anisotropy may introduce a coupling between such processes, a frequent
situation in biological problems, as in active transport of ions or molecules
across cellular or tissue membranes, or along lengthy microscopic filaments.
These kinds of transport are attributed to enzymes, which act as molec-
ular engines. The study of microscopic machines, natural or artificial, is at
the frontiers of nowadays science. In the past, the calculation of the max-
imum efficiency of heat engines has represented a decisive contribution to
equilibrium thermodynamics. Similarly, the determination of the efficiency
of energy transfer between coupled processes may be considered as one of the
main achievements of linear irreversible thermodynamics.
As an illustration, consider the active transport of sodium through a cel-
lular membrane. Cells at rest state have an internal electrical potential of
−70 mV as compared to the external potential, taken as reference (0 mV).
Since the internal concentration of sodium is low as compared to the exter-
nal concentration, sodium has a natural tendency to flow towards the interior
of the cell. Now it is observed that the cell pumps out as much sodium as the
ingoing one but, of course, this transport cannot be spontaneous, because
it takes place against the natural tendency of matter to flow from regions
104 4 Chemical Reactions and Molecular Machines
of higher to lower electrochemical potential. This transport is made at the
expenses of the energy supplied by the hydrolysis of ATP, the usual fuel in
biological organisms.
We are in presence of a coupling between a chemical reaction, characterized
by an affinity / and a chemical flux w, and a transport of matter; it is shown
in Sect. 2.7 on matter diffusion that the latter is described by a force, the
difference of chemical potential ∆µ between the inner and the outer regions,
and a flux of matter (sodium in the present case) J. The corresponding rate
of dissipation is given by

s
= /w +J∆µ, (4.40)
with /w > 0, because the hydrolysis of ATP takes place along the natural
tendency given by / > 0 and w > 0, whereas J∆µ < 0, because the flow
of sodium (J < 0) is opposite to the direction of ∆µ. We are faced with
an example where a reaction with positive entropy production allows for the
occurrence of a simultaneous process, with a negative contribution to the
entropy production.
The phenomenological laws describing the kinetics of the reaction and the
rate of transport are given by
w = L

11
/+L

12
∆µ, (4.41)
J = L

21
/+L

22
∆µ, (4.42)
with L

12
= L

21
according to Onsager’s reciprocal relations. The coefficient
L

21
in (4.42) is responsible for the coupling between chemical reaction and
mass transport.
For instance, the experimental values of the phenomenological coefficients
for the transport across the membrane of a frog bladder are the following:
L

22
= 104, L

12
= 5.40, and L

11
= 0.37 in units of mol
2
cm
−2
s
−1
kcal
−1
. The
coupling parameter q defined by (4.35) is given by q = 0.86 and the maximum
efficiency calculated from (4.39) is η
max
= 33%.
In so-called linear motors, i.e. motors which transport matter along lengthy
linear macromolecules, as kinesin along microtubules, the resistance is due
to viscous friction rather than to a jump of the electrochemical potential. In
Box 4.3, the modus operandi of such a molecular motor is presented. The
expression for the dissipated energy is then

s
= /w +vF, (4.43)
v being the velocity of the machine along the filament and F the viscous
resistance. The phenomenological laws are, in the linear approximation,
w = L
11
/+L
12
F, (4.44)
v = L
21
/+L
22
F, (4.45)
4.4 Chemical Reactions and Mass Transport: Molecular Machines 105
with the coefficient L
22
inversely proportional to the viscosity of the medium.
Since molecular machines are of very small dimensions, the energy related to
their several steps is comparable or smaller than the thermal noise. Their
behaviour is therefore not deterministic but stochastic, in contrast with that
of macroscopic engines. The understanding of the relations between thermal
fluctuations and the behaviour of molecular machines is a subject of current
interest, as explained in Box 4.4.
Box 4.3 Biological Molecular Motors with Three Configurations
A model for chemically driven transport may be based on a modification
of the triangular reaction scheme of Sect. 4.2.2. An abstract illustration is
provided by the following molecular motor running through three different
configurations, denotedas M
1
, M
2
, andM
3
(there wouldbe no difficultyto add
more configurations), and working in a cyclic way. We assume that the motor
is fuelled with ATP and hydrolyses it, changing the configuration at each
chemical step. Thus, instead of a closed reaction triangle, one has the scheme
(M
1
) + ATP →(M
2
) ATP, (4.3.1)
(M
2
) ATP →(M
3
) ADP + P
i
, (4.3.2)
(M
3
) ADP →(M
1
) + ADP. (4.3.3)
To be explicit, configuration 1 bounds one molecule of ATP, which pro-
duces a change to configuration M
2
. In a second step, ATP is hydrolysed:
the inorganic phosphate P
i
is liberated and the macromolecule changes to
configuration M
3
, with ADP still bound to it. Finally ADP is liberated
from configuration M
3
and the motor returns to the initial configuration
M
1
, where the cycle may start again. At each cycle, the motor does some
work: either it advances a step along a macromolecule (as for instance ki-
nesin along tubulin or myosin along actin) or it translocates some molecule
(as for instance in membrane ionic pumps). For more biological details, the
reader is referred, for instance, to Nelson (2004).
Finally, the triangular scheme would be
106 4 Chemical Reactions and Molecular Machines
It is assumed that the kinetic rate constants of reactions (4.3.1)–(4.3.3) and
k
−1
, k
2
, and k
3
in the triangular scheme do not depend on the ATP nor
ADP concentration, whereas k
1
= k
(0)
1
[ATP], k
−3
= k
(0)
−3
[ADP], and k
−2
=
k
(0)
−2
[P] (where k
(0)
1
, k
(0)
−3
, and k
(0)
−2
are constant values) are pseudo-first-order
rate coefficients into which the concentrations of ATP, ADP, and P
i
are
incorporated. This makes an essential difference with the triangular reaction
scheme presented in Sect. 4.2.2. Indeed when the hydrolysis reaction
ATP →ADP + P
i
(4.3.4)
is in equilibrium, the cycle has not net motion; however, when the affinity
of this reaction is positive, the cycle will turn in the direction 123, whereas
if the affinity is negative, it will move in the opposite direction 321. Such
kind of model machine may both hydrolyse ATP and produce some work
pumping ions from lower to higher electrochemical potential (active trans-
port) or, in the reverse way, it may draw some energy from the ions crossing
it from higher to lower electrochemical potential and produce ATP.
Box 4.4 Brownian Motors
Molecular motors are sensitive to thermal noise which might be used, to
some extent, to produce useful work, provided some structural and ther-
modynamic conditions are satisfied. Instead of performing a simple random
Brownian motion, these motors may rectify, to some extent, the thermal
noise. The modelling of such molecular motors has been much developed
since 1990 (Magnasco 1994; Astumian and Bier 1994; J¨ ulicher et al. 1997;
Astumian 1997) (Fig. 4.3).
Fig. 4.3 Transport driven by a chemical reaction (Astumian 1997)
4.4 Chemical Reactions and Mass Transport: Molecular Machines 107
As an example, we analyse a simple model proposed by Astumian (1997)
for chemically driven transport. The author studies the one-dimensional
motion of a charged particle along a periodic lattice of dipoles arranged
head to tail (as found in many biological macromolecules, like tubulin or
actin). The interaction between the particle and the linear arrangement
of dipoles is assumed to be merely electrostatic (in biology, the couplings
may also be conformational). The potential energy profile has essentially
the form of a sawtoothed function as that shown in Fig. 4.3 (if the particle
is negatively charged, the minima of the sawtooth will correspond to the
positive ends of the dipoles and the maxima to the negative ends). Because
of thermal noise, the particle undergoes Brownian motion along the line,
and occasionally has enough energy to pass one of the neighbouring barriers.
However, despite the anisotropy of the potential profile, the probabilities of
moving to the right or to the left are equal.
Assume, instead, that the particle catalyses a chemical reaction SH
S

+H
+
, with the product particles being charged; S designates some sub-
stract species. Now, the amplitude of the potential will depend on whether
H
+
or S

are bound to the particle, resulting in a coupling between the
chemical reaction and the diffusion of the particle along the line. However,
when the particle is bound only to H
+
(assume that the ionized substract
S

has been expelled to the bulk), the particle becomes uncharged and there
is no longer a barrier opposing its motion. Therefore, the particle may move
along the line. Once H
+
is liberated, the charge of the particle constitutes
again a barrier. However, it may be shown that an asymmetric potential
whose barriers are fluctuating in height gives raise to a directional motion
of particles. The corresponding flux is given by (Astumian and Bier 1994)
J ≈
γk
B
T
4
_
erf
_
α
2
ζ
_
−erf
_
1 −α
2
ζ
__
,
where α is the asymmetry parameter of the potential, and ζ ≡ (γL
2
/D)
1/2
while γ, L, and D are typical parameters of the model. If the saw-
toothed potential is symmetric (α = 0.5), J = 0 and there is no net
flow.
To summarize, it turns out that molecular machines can convert free en-
ergy into motion under the conditions that this mechanism is out of equi-
librium and structurally asymmetric. Moreover, molecular engines cannot
increase their speed without limit as the energy supply is increased, because
the speed saturates at some limiting value. This means that linear transport
relations are not indefinitely valid, but break down for high values of fluxes
and forces.
108 4 Chemical Reactions and Molecular Machines
4.5 Autocatalytic Reactions and Diffusion:
Morphogenesis
Another consequence of the coupling between chemical reactions and diffu-
sion is the occurrence of patterns, i.e. spatial inhomogeneities, in systems far
from equilibrium. Although this topic is studied in detail in Chap. 6, we find
it interesting to present here an introductory version based on a model pro-
posed by Turing (1952), who settled the foundations of morphogenesis, i.e. the
transition of an originally homogeneous system into an inhomogeneous one.
This is a characteristic feature of non-equilibrium thermodynamics, because
near equilibrium, transport phenomena (matter diffusion, thermal conduc-
tion) tend to make the system spatially homogeneous in the long run.
Consider two fluid subsystems at the same pressure and temperature,
which are separated by a permeable membrane. In each of them, the fol-
lowing autocatalytic reaction will take place between substances A and X
A + 2X →3X. (4.46)
In this reaction, a substance A is transformed into the substance X, at a rate
which depends on the concentration of X. The molar concentrations of X
in both subsystems, ˜ c
X
1
and ˜ c
X
2
, respectively, tend to become equal due to
diffusion of X across the membrane. The diffusion flux of X from subsystem
1 to subsystem 2 is given by
J
1→2
= α(˜ c
X
1
− ˜ c
X
2
), (4.47)
where α is a constant proportional to the permeability and the area of the
membrane.
We will show that, under some conditions, the difference of concentrations
˜ c
X
between both subsystems will increase because of the autocatalytic re-
action (4.46). First, we note that the rate of production of substance X is
higher where the molar concentration ˜ c
X
is higher, as it follows from the mass
action law
d˜ c
X
dt
= k˜ c
A
(˜ c
X
)
2
, (4.48)
where k is a positive kinetic constant.
Taking diffusion into account, the evolution of ˜ c
X
1
as a function of time in
subsystem 1 is given by
d˜ c
X
1
dt
= −α(˜ c
X
1
− ˜ c
X
2
) +k˜ c
A
(˜ c
X
1
)
2
, (4.49)
whereas the evolution of ˜ c
X
2
in subsystem 2 is
d˜ c
X
2
dt
= −α(˜ c
X
1
− ˜ c
X
2
) +k˜ c
A
(˜ c
X
2
)
2
. (4.50)
By subtracting both equations, one obtains for the evolution of the difference
of concentrations ˜ c
X
1
− ˜ c
X
2
,
4.6 Problems 109
d(˜ c
X
1
− ˜ c
X
2
)
dt
= −2α(˜ c
X
1
− ˜ c
X
2
) +β˜ c
A
_
(˜ c
X
1
)
2
−(˜ c
X
2
)
2
¸
. (4.51)
This may still be written as
d(˜ c
X
1
− ˜ c
X
2
)
dt
= −2α(˜ c
X
1
− ˜ c
X
2
) +k˜ c
A
(˜ c
X
1
+ ˜ c
X
2
)(˜ c
X
1
− ˜ c
X
2
). (4.52)
This result indicates that the right-hand side of (4.52) becomes positive,
thus implying an increase of the difference of concentrations, when the molar
concentration of A becomes higher than the critical value
(˜ c
A
)
c
=

k
(˜ c
X
1
+ ˜ c
X
2
). (4.53)
Note that, since reaction (4.46) is irreversible, i.e. it only goes to the right, the
concentration of A corresponding to equilibrium is zero. In the problem under
study, ˜ c
A
remains constant because it is supplied from the outside at the same
rate in both subsystems. If the supply rate is sufficiently high in such a way
that the concentration of A exceeds the critical value, the concentration ˜ c
X
will tend to be higher in one subsystem than in the other and will contribute
to the reinforcement of the non-homogeneity of the system.
This mechanism has gained much interest in biology. Indeed, an actual
problem in biology is the development of an embryo from the first fertilized
cell of the individual and this process implies cell differentiation at the var-
ious stages of development. Though all the cells of the individual contain
the same information in their DNA, the parts of this information which are
really read are not the same in all cells, but they depend on control proteins
that determine which genes are transduced into the corresponding proteins.
In the above model, one could interpret X as the control molecule determining
the reading of some gene. When the system is far enough from equilibrium,
the inhomogeneity in the value of ˜ c
X
makes that, in one part of the organism
where ˜ c
X
is higher than some critical value, a given gene is expressed, whereas
in the other regions, where ˜ c
X
is lower, the gene is not expressed.
4.6 Problems
4.1. Michaelis–Menten’s relation for enzymatic reactions. An important type
of biochemical reactions are reactions changing a substrate S into a product P,
catalysed by an enzyme E, through the intermediate formation of an enzyme–
substrate complex ES. The reaction may be expressed as
E + S ES →E + P.
The first step is reversible and the second one is irreversible. The correspond-
ing kinetic constants are k
1
and k
−1
for the forward and backward steps of
110 4 Chemical Reactions and Molecular Machines
the first reaction, and k
2
for the (forward) step of the second reaction. The
velocity of reaction w ≡ d˜ c
P
/dt = −d˜ c
S
/dt obeys the Michaelis–Menten
expression
w = w
max
˜ c
S
K
M
+ ˜ c
S
,
where w
max
is the saturation velocity and K
M
a characteristic constant indi-
cating the value of the substrate molar concentration ˜ c
S
for which the velocity
is half the maximum value. (a) Show that, according to the mass action law,
the value of the concentration of the complex ES in the steady state (for
which the enzyme concentration ˜ c
E
is constant) is given by
˜ c
ES
=
k
1
˜ c
S
k
−1
+k
2
+k
1
˜ c
S
.
(b) It follows from the mass action law that the velocity of the reaction is
w = d˜ c
P
/dt = k
2
˜ c
ES
. Combine this result with the previous one and show
that Michaelis–Menten expression is recovered. Identify w
max
and K
M
in
terms of the kinetic constants k
1
, k
−1
, k
2
, and of the enzyme concentration
˜ c
E
. Discuss the dependence of w
max
with ˜ c
E
and the relevance of the fact
that the last step is totally irreversible (i.e. the backward reaction is lacking).
(c) Assume that for a given enzymatic reaction w
max
= 0.085 mMs
−1
and
K
M
= 6.5 mM. Obtain the velocity w for ˜ c
S
= 2.5, 10, and 25 mM.
4.2. Application of Michaelis–Menten expression to simple molecular motors.
The kinetics of simple, tightly coupled chemical motors with at least one ir-
reversible step is expected to be described by a Michaelis–Menten expression,
discussed in the previous problem. Assume, for instance, the following scheme
motor(x) + ATP ATP motor →motor(x + ∆x) + ADP.
In this reaction, the hydrolysis of ATP gives the energy necessary for moving
the motor from a position x to x +∆x along some filament, the motor could
be kinesin and the corresponding filament could be a microtubule, in which
case ∆x = 8 nm. The speed with which the motor moves along the filament
can be cast in the form
v = v
max
(F)
˜ c
ATP
K
M
(F) + ˜ c
ATP
,
wherein the maximum velocity v
max
and the characteristic constant K
M
de-
pend on the load force acting on the motor. Indeed, it has been found that,
for kinesin along microtubules, the numerical values are v
max
≈ 815 nms
−1
and K
M
≈ 90 µM at F = 1 pN, and v
max
≈ 400 nms
−1
and K
M
≈ 300 µM at
F = 5.5 pN. (a) Compare the velocity at ˜ c
ATP
= 1, 10, and 100 mM, under
both load charges. (b) Compare the useful power delivered by the motor,
given by Fv, in the several situations mentioned above.
4.3. Conditions of maximum efficiency. Consider a transport process across
a membrane, when the affinity of the hydrolysis of ATP is /
1
and assume
4.6 Problems 111
that the transported substance may be considered as ideal and uncharged.
Show that the ratio of internal and external concentrations of the substance,
under the conditions of optimum free-energy exchange between ATP energy
and transfer, is given by
ln
c
in
c
out
=
/
1
RT
(1 −q
2
)
1/2
−1
qZ
,
the coupling coefficients q and Z are defined in Sect. 4.3, namely q ≡
L

12
/(L

11
L

22
)
1/2
and Z ≡ (L

11
/L

22
)
1/2
, L

ij
being the phenomenological
coefficients of the linear constitutive equations relating the velocity of the
chemical reaction (taken as flux 1) and the rate of active transport (taken as
flux 2) with the corresponding conjugated thermodynamic forces.
4.4. Kinetics of triangular reaction. An example of a triangular reaction
scheme is provided by the triangular catalytic isomerization of the three
butenes C
3
H
8
(namely, forms (A), (B), and (C) shown below), was stud-
ied by W. O. Haag and H. Pines (J. Am. Chem. Soc. 82 (1960) 387 and
2488). They obtained for the corresponding kinetic constants in the presence
of Al
2
O
3
catalyst: k
1
/k
3
= 2.4, k
−2
/k
−1
= 1.0, and k
−3
/k
2
= 0.4. Analyse
whether these results are compatible with the statement that the global equi-
librium in the triangular reactions corresponds to equilibrium in each of the
three individual reactions (see Chartier et al. 1975).
(A) H
2
C == CH — CH
2
— CH
3

( ) B CH — CH
3
(cis)
CH — CH
3
( ) CH — CH
3
C
CH
3
— CH
(trans)
4.5. Autocatalytic reactions and diffusion. Discuss the instability problem
analysed in Sect. 4.5 when one assumes that A+X →2X instead of A+2X →
3X, as in (4.46).
4.6. Entropy production and degree of coupling in kinesin motor. Kinesin, a
typical molecular machine which moves along microtubules, takes on average
one 8-nm step, corresponding to one monomer of tubulin, every 10 ms. A
single ATP molecule is hydrolyzed per step. The chemical energy relased by
ATP hydrolysis is typically 20 k
B
T, and the motor uses 12 k
B
T with each step
(Bustamante et al. 2005). a) Compute the power and the force done by the
kinesin, and the hydrolysis velocity reaction. b) What is the efficiency of the
112 4 Chemical Reactions and Molecular Machines
machine? c) Evaluate the entropy production of the machine in terms of k
B
per second. d) If this efficiency is the maximum one, evaluate from equation
(4.39) the degree of coupling between ATP hydrolysis and kinesin motion,
according to the scheme provided in (4.44) and (4.45) for linear motors.
4.7. Bacterial flagellar motor. Swimming bacteria are usually propelled by
a rotary motor embedded in the cell envelope which drives rotating flagellar
filaments. These motors are typically powered by the inflow of protons across
the cell membrane, and have very interesting properties which confer bacteria
with the possibility of chemiotaxis, i.e., swimming towards higher nutrient
concentrations. We simplify the complexity of this motor and assume that
the power delivered by the protons, namely Power = J∆µ, is applied with
unit efficiency to propel the cell. Here J is the proton flux across the motor
(number of protons per second) and ∆µ is the socalled protonmotive force,
defined as
∆µ = ∆V + 2.303(k
B
T/e)∆pH
∆V being the membrane potential (difference of potential across the mem-
brane) and ∆pH being ∆pH = log(c
H+,out
/c
H+,in
), with c
H+
being the proton
concentration. Assume that a bacterium is a sphere of 1 µm of radius, and
its friction coefficient is given by the Stokes law, i.e. ζ = 6πηr, with r the
radius of the cell and η the viscosity of the surrounding fluid. Assume that
when the membrane potential is −80 mV (inside is negative) and ∆pH = 0,
the bacterium moves at 25 µm/s. a) Find the proton flux necessary to propel
the cell in water (η = 10
−3
N s m
−2
). b) Find the electrical resistance of the
motor. c) Evaluate the velocity of the cell when it swims in a solution with
pH = 6 (assume that inside the cell pH = 7). d) Compare the total entropy
production due to viscous dissipation in both cases.
4.8. Degree of coupling and maximum power. In expression (4.39) was given
the maximum efficiency in terms of the degree of coupling q of two coupled
chemical reactions. In some circumstances, however, as for instance in hunting
or fleeing, so important for survival, the organism is interested in obtaining
the maximum power instead of the maximum efficiency. a) Write the power
output −w
2
A
2
in the normalized form −w
2
A
2
/L
11
A
2
1
and show that it may
be expressed as −Zx(Zx + q); b) Show that the corresponding efficiency at
the value of q for which the power output is maximum is
η
max power
= (q
2
/4)[1(q
2
/2)]
−1
.
c) Compare this value with the maximum efficiency.
4.9. Effects of the breaking of detailed balance on the Onsager relations. As-
sume that in the triangular cyclical set of chemical reactions presented in
Section 4.2.2, detailed balance was broken, i.e., instead of having w
1
= w
2
=
w
3
= 0 one had w
1
= w
2
= w
3
= w. How would the phenomenological coeffi-
cients L
12
and L
21
appearing in the constitutive equations (4.25) be related?
(i.e. express L
12
−L
21
in terms of w).
Chapter 5
Finite-Time Thermodynamics
Economy, Ecology, and Heat Engines
One of the motivations of the early developments of thermodynamics was
the optimization of heat engines, as steam engines, which transform partially
heat into work, and which led to the first industrial revolution between the
end of the eighteenth century and the beginning of the nineteenth century.
This provides an interesting illustration of how an engineering problem cou-
pled with economical needs stimulated the occurrence of a new fundamental
science. Nowadays, we live in a similar situation: we are experiencing an in-
dustrial revolution based on information processing and miniaturized engines
and, on the other side, we are conscious of the need to incorporate ecological
restrictions into the economical progress. To convey some flavour of these
problems, we present in this chapter applications of non-equilibrium ther-
modynamics to heat engines working with a finite non-vanishing rate. This
approach is referred to as finite-time thermodynamics and received a strong
impetus during the last quarter of the twentieth century.
In 1824, Carnot arrived to the conclusion that the efficiency of a heat
engine operating between a hot and a cold heat reservoir is maximum in
reversible processes, and that it depends only on the temperature of the
two reservoirs, but not on the working substance. Here, Carnot’s ideas are
extended by including considerations not only on the efficiency but also on
the power developed by the engine when the cycle is performed in a finite
time. This issue is of importance from the economical perspective, because
what the society needs is an amount of work in a finite time, rather than
work supplied at an infinitesimal small rate, as in reversible engines.
Though the mutually conflicting demands in the maximization of efficiency
and power are especially illustrative of the conceptual problems dealt with
in this chapter, we also compare other optimization criteria, as for instance,
those based on the minimization of entropy production, and on maximization
of the power minus the product of temperature and entropy production.
This diversity of criteria illustrates the conceptual subtleties arising in finite-
time thermodynamics. Furthermore, the time constraints must be included
in realistic models of engines to optimize their design and performance.
113
114 5 Finite-Time Thermodynamics
Finite-time thermodynamics deals with a very wide collection of situa-
tions. In the present chapter, we only present some simple and pedagogical
illustrations, based on Carnot’s engine, but many other problems have been
investigated in finite-time thermodynamics as, for instance, how to optimize a
thermodynamic cycle when its total duration is fixed. In this case, the several
branches of the cycle may be performed at different rates, and the problem is
to find the rates of each process in such a way that some optimization crite-
rion (power output maximization, for instance) is satisfied. Finite-time ther-
modynamics is not limited to Carnot cycle, but it encompasses many other
cycles (Otto, Brayton, Stirling, Diesel, etc.), and it is concerned not only
with ideal gases but also with other kinds of fluids, as polytropic fluids. Par-
ticular studies have been devoted to refrigerators, heat pumps, engines with
heat regeneration, several coupled engines, and have been applied to solar
engines, photovoltaic cells, or to the evaluation of the maximum efficiency of
the conversion from solar energy to wind energy in the planet. In short, finite-
time thermodynamics allows us to understand in a simple way the essential
role of time constraints on the thermodynamic cycles (Salamon et al. 1980;
Ma 1985; Salamon and Sieniutycz 1991; Bejan 1996; Hoffmann et al. 1997).
This is a clear illustration of the importance of including non-equilibrium
thermodynamics both in the assessment and design of heat engines as well
as in the understanding of some natural phenomena.
5.1 The Finite-Time Carnot Cycle
Let us briefly recall the essentials of the classical analysis of Carnot’s engine.
The latter is operating between two heat reservoirs at temperatures T
1
and T
2
(T
1
> T
2
), it takes an amount of heat Q
1
from the hot one, transforms a part
of it into work W, and delivers the remaining heat Q
2
to the cold reservoir.
Since the engine operates cyclically, Q
1
, Q
2
, and W refer, respectively, to the
amounts of heat, absorbed and delivered, and the work carried out by the
engine during one cycle.
The internal energy of the engine does not change during one cycle, because
it is a state function, so that the first law takes the form
Q
1
= W +Q
2
. (5.1)
If we ask about the efficiency η of the cycle – namely the work produced W
divided by the heat supplied Q
1
to the engine – one has
η ≡
W
Q
1
= 1 −
Q
2
Q
1
. (5.2)
Kelvin–Planck formulation of the second law states the impossibility of build-
ing a cyclical heat engine, which transforms heat completely into work. In-
deed, the maximum value of η is limited by the minimum value of the heat
5.1 The Finite-Time Carnot Cycle 115
Q
2
delivered to the cold reservoir. Therefore, to evaluate the maximum effi-
ciency, the second law must be taken into account. Accordingly, the global
balance of entropy is never negative, i.e.
∆S
tot
= ∆S
1
+ ∆S
2
+ ∆S
engine
≥ 0, (5.3)
where ∆S
1
, ∆S
2
, and ∆S
engine
are the variations of entropy, respectively, in
the reservoirs 1 and 2 and in the engine during one cycle. Since the engine
operates cyclically, and recalling that the entropy is a function of state, one
has ∆S
engine
= 0. Moreover, the temperature of each reservoir remains con-
stant, so that, according to Clausius’ definition of entropy, the variations of
their respective entropies are ∆S
1
= −Q
1
/T
1
and ∆S
2
= Q
2
/T
2
. Therefore,
(5.3) becomes
∆S
tot
= −
Q
1
T
1
+
Q
2
T
2
= ∆
irr
S ≥ 0. (5.4)
The equality sign corresponds to reversible processes and ∆
irr
S stands for
the entropy produced during one cycle by irreversible processes (as for in-
stance thermal resistance or mechanical and hydrodynamical friction). After
combining (5.2) and (5.4), it follows that the efficiency obeys the inequality
η ≤ 1 −
T
2
T
1
. (5.5)
This expression shows that the efficiency is maximum when the engine op-
erates reversibly: indeed, in this case the relation (5.5) becomes an equality,
providing the so-called Carnot’s efficiency, depending only on the temper-
ature of the sources but not on any detail of the working substance or the
engine itself.
However, in order that the engine operates reversibly, one must proceed
infinitely slowly, with the result that the power delivered by the reversible
engine is in fact zero. This is of no practical interest because, in actual sit-
uations, a non-vanishing power is desirable, which implies to obtain a finite
amount of work in a finite time. Anyway, one should avoid spending unre-
alistic efforts to improve engines beyond their actual physical limits; in that
respect, the result (5.5) indicates that a way to enhance the efficiency is to
raise as much as possible the temperature of the hot reservoir and to reduce
the temperature of the cold reservoir.
A cycle carried out in a finite time means a reduction of the efficiency, and
therefore one is faced with an alternative: maximization of efficiency or max-
imization of power. To study this problem, we will follow in this introductory
section the didactical example proposed by Curzon and Ahlborn (1975).
5.1.1 Curzon–Ahlborn’s Model: Heat Losses
Curzon and Ahlborn assume that the rate of heat exchanged between the
working substance at a temperature T

1
and the heat reservoir at a tempera-
ture T
1
is described by Newton’s law of heat transport, namely
116 5 Finite-Time Thermodynamics
Fig. 5.1 Sketch of the Curzon–Ahlborn’s model of heat engine with finite thermal
resistance between the working fluid and the heat reservoirs
dQ
1
dt
= α
1
(T
1
−T

1
), (5.6)
where the coefficient α
1
depends on the thermal conductivity of the wall
separating the working fluid from the source and on geometrical factors, as
the area and the width of the wall. Analogously, the rate of heat delivered to
the cold source is
dQ
2
dt
= α
2
(T
2
−T

2
), (5.7)
where α
2
plays a role analogous to α
1
in the heat exchange between the
working fluid and the cold reservoir. A sketch of the Curzon–Ahlborn’s model
is provided in Fig. 5.1.
It follows from (5.6) and (5.7) that the times required for the engine to
absorb an amount of heat Q
1
and to deliver the heat Q
2
are, respectively,
given by
t
1
=
Q
1
α
1
(T
1
−T

1
)
, t
2
=
Q
2
α
2
(T

2
−T
2
)
. (5.8)
By writing these expressions, we have assumed that the temperatures T

1
and T

2
of the working substance remain constant during the heat exchange
process. Furthermore, for the sake of simplicity, it is admitted that the time
spent in the adiabatic parts of the cycle is much shorter than the time required
for absorbing or delivering the heat. Physically, this implies the absence of
frictional losses and inertial effects (e.g. one works with a frictionless piston
of zero mass). In this case, the total duration of the cycle is
t
cycle
≈ t
1
+t
2
. (5.9)
This theoretical framework has the advantage of allowing explicit calculations
of all the quantities related to the way the heat engine is operating, and to
5.1 The Finite-Time Carnot Cycle 117
introduce explicitly the notion of time. In comparison with reversible ther-
modynamics, one makes a step forward by capturing more realistic features
of the heat engine.
The power P delivered by the engine during one cycle is the ratio of the
work W that is performed and the total duration of the cycle, i.e.
P =
W
t
cycle
. (5.10)
Now, we want to obtain both efficiency and power as a function of the dura-
tion of the cycle. To do this, we need the expression of the entropy produced
during a cycle, which is given by (5.4). On the other hand, we know from
Chap. 2 that, when an amount of heat Q flows from a system at temperature
T to another at temperature T

, the amount of entropy produced is

irr
S = −Q
_
1
T

1
T

_
. (5.11)
The total entropy generated during one cycle is assumed to result exclusively
from irreversible heat exchanges between the reservoirs and the working sub-
stance, all other possible sources of irreversibility, as for instance viscosity,
are neglected for the moment. In the above problem, one has

irr
S = −Q
1
_
1
T
1

1
T

1
_
−Q
2
_
1
T

2

1
T
2
_
. (5.12)
Making use of (5.4), one obtains

Q
1
T

1
+
Q
2
T

2
= 0. (5.13)
Compared to the equivalent result (5.4), wherein the temperatures T
1
and T
2
of the sources appear, in (5.13) the temperatures of the working fluid dur-
ing the isothermal expansion and compression are coming out. The analogy
between (5.13) and (5.4) for the reversible situation (i.e. with ∆
irr
S = 0)
underlines that the working fluid itself is behaving in a reversible way, and
that the only source of irreversibility is the heat transfer between systems at
different temperatures (i.e. between the working fluid and the corresponding
heat reservoirs). This is called an endoreversible heat engine (de Vos 1992;
Andresen 1996; Hoffmann et al. 1997), because irreversibilities find their ori-
gin in the contact with the sources but not in internal causes. In particular,
for a given compression ratio V
max
/V
min
of an ideal gas and using the first
law of thermodynamics, the values of Q
1
and Q
2
are
Q
1
= RT

1
ln(V
max
/V
min
), Q
2
= RT

2
ln(V
max
/V
min
). (5.14)
We now study the conditions under which the power becomes a maximum.
The work per cycle W is equal to the difference Q
1
−Q
2
, and will be the larger
118 5 Finite-Time Thermodynamics
Fig. 5.2 Normalized efficiency and power output of the heat engine as a function
of the inverse of the duration of the cycle
when the higher is Q
1
and the lower is Q
2
. According to (5.13), this requires
to increase T

1
and to reduce T

2
, because Q
1
/Q
2
= T

1
/T

2
. But if T

1
increases
and T

2
decreases, the rate of heat exchanged with the sources, expressed by
(5.6) and (5.7), diminishes. Under these circumstances, one has slow cycles
with high efficiency but small power: each cycle produces much work but one
cycle takes a long time. If the rate of heat is increased by lowering T

1
and
rising T

2
, efficiency is reduced: each cycle produces less work but it takes a
shorter time. When the rate is too high, the work per cycle becomes very low
in such a way that the power decreases and becomes eventually zero. Between
these extreme situations with a vanishing power, there is another one with
a maximum power output, as sketched in Fig. 5.2. In this figure, the zero
values of power output correspond either to very low speed per cycle (high
efficiency but small power) or to very high speed per cycle (low efficiency and
low power).
To determine the values of T

1
and T

2
corresponding to the maximum power
output, at fixed values of the temperatures T
1
and T
2
of the heat reservoirs,
let us write explicitly the expression for the power output, namely
P(T

1
, T

2
) =
Q
1
−Q
2
t
cycle
=
Q
1
−Q
2
Q
1
α
1
(T
1
−T

1
)
+
Q
2
α
2
(T

2
−T
2
)
. (5.15)
Eliminating Q
1
and Q
2
in terms of T

1
and T

2
by using (5.13), one obtains
P(T

1
, T

2
) =
T

1
−T

2
T

1
α
1
(T
1
−T

1
)
+
T

2
α
2
(T

2
−T
2
)
. (5.16)
After introducing the variables x
1
≡ T
1
− T

1
and x
2
≡ T

2
− T
2
, (5.16) may
be rewritten in the more compact form
5.1 The Finite-Time Carnot Cycle 119
P(x
1
, x
2
) =
α
1
α
2
x
1
x
2
(T
1
−T
2
−x
1
−x
2
)
α
2
T
1
x
2

1
T
2
x
1
+ (α
1
−α
2
)x
1
x
2
. (5.17)
The conditions for P being extremum are
∂P(x
1
, x
2
)
∂x
1
= 0,
∂P(x
1
, x
2
)
∂x
2
= 0. (5.18)
This leads to two equations for the values of x
1
and x
2
corresponding to the
maximum power,
α
2
T
1
x
2
(T
1
−T
2
−x
1
−x
2
) = x
1

2
T
1
x
2

1
T
2
x
1
+ (α
1
−α
2
)x
1
x
2
), (5.19)
α
1
T
2
x
1
(T
1
−T
2
−x
1
−x
2
) = x
2

2
T
1
x
2

1
T
2
x
1
+ (α
1
−α
2
)x
1
x
2
). (5.20)
It is verified that
x
2,max
=
_
α
1
T
2
α
2
T
1
_
1/2
x
1,max
(5.21)
with
x
1,max
=
T
1
−(T
1
T
2
)
1/2
1 + (α
1

2
)
1/2
, x
2,max
=
(T
1
T
2
)
1/2
−T
2
1 + (α
2

1
)
1/2
. (5.22)
In particular, it is found from (5.21) and (5.22) that
T

1
= CT
1/2
1
, T

2
= CT
1/2
2
, C ≡

1
T
1
)
1/2
+ (α
2
T
2
)
1/2
α
1/2
1

1/2
2
. (5.23)
Introducing (5.22) in (5.17), we obtain for the maximum power output
P
max
=
α
1
α
2

1/2
1

1/2
2
)
2
(T
1/2
1
−T
1/2
2
)
2
. (5.24)
The efficiency as a function of x
1
and x
2
is
η(x, x) = 1 −
T

2
T

2
= 1 −
T
2
+x
2
T
1
−x
1
. (5.25)
Bringing it in the expressions (5.22) yields the important relation
η
max power
= 1 −
_
T
2
T
1
_
1/2
. (5.26)
It is worth to stress that, in contrast with Carnot’s result (5.5), the efficiency
at maximum power output relates to the square root of the ratio of temper-
atures. Both relations (5.24) and (5.26) are of high interest, especially the
second one, which is seen to depend only on the values of the temperatures
of the reservoirs, in contrast to the first one, which is also function of the
heat transfer coefficient introduced in (5.6) and (5.7).
120 5 Finite-Time Thermodynamics
As an illustration, consider a power station working, for instance, between
heat reservoirs at 565 and 25

C and having an efficiency of 36%. We want to
evaluate this power station from the thermodynamic point of view. It follows
from (5.5) that the Carnot’s maximum efficiency is 64.1%; our first opinion
on the quality of the power station would be rather negative if Carnot’s
efficiency is our standard for evaluation. However, according to (5.26), its
efficiency at maximum power is 40%. Thus, if the objective of the power
station is to work at maximum power output, it is seen that the efficiency
is not bad compared to the 40% efficiency corresponding to this situation.
We can therefore conclude that, to have realistic standards for evaluation
of an actual heat engine, one needs to go beyond Carnot’s efficiency and to
incorporate finite-time considerations.
We have stressed that (5.26) is independent of the heat transfer coefficients,
and therefore it is rather general. However, its generality is not so universal as
Carnot’s result (5.5), because it has been derived by assuming that thermal
conduction is the only source of irreversibility, thus neglecting other possible
dissipative processes, and also because it depends on the specific form adopted
for the heat transfer law. For instance, instead of Newton’s law taken in (5.6)
and (5.7), one could assume a transfer law of the form
dQ
1
dt
= β
1
_
1
T

1

1
T
1
_
,
dQ
2
dt
= β
2
_
1
T
2

1
T

2
_
. (5.27)
In this case, it may be shown that the efficiency at the maximum power is
given by
η(at P
max
) =
1 + (β
1

2
)
1/2
2 +
_
1 +
T
2
T
1
_

1

2
)
1/2
_
1 −
T
2
T
1
_
. (5.28)
Compared to (5.26), this expression depends now explicitly on the heat trans-
fer coefficients. Several other heat transfer laws have been proposed in the
literature (e.g. Hoffmann et al. 1997).
5.1.2 Friction Losses
In the previous examples, the presence of internal dissipation has been ig-
nored. Dissipation could be, for instance, produced by friction of the piston
along the walls, heat leaks due to imperfect insulation of the reservoirs, in-
ternal losses in the working fluid, such as turbulence or chemical reactions.
A simple way to introduce friction is to assume that the total force acting on
the piston is
F = pA−fv, (5.29)
where pA is the force exerted by the gas on the piston of area A and fv
represents a frictional viscous force, v being the velocity of displacement
5.1 The Finite-Time Carnot Cycle 121
of the piston and f the friction coefficient. This model was investigated by
Rebhan (2002), who assumed that heat transfer effects are arbitrarily fast and
that the only source of dissipation is friction. This is the situation analysed
below.
During a displacement dx, the work performed by the piston will be
dW = p dV −fv
2
(t)dt, (5.30)
and the work corresponding to one cycle is
W = Q
1
−Q
2
−f
_
c
v
2
(t)dt, (5.31)
the integration being performed over the duration of one cycle. The compu-
tation of the last term depends on the time dependence of the velocity. Let
us qualitatively write
W = Q
1
−Q
2

4fδD
2
τ
, (5.32)
where 2D is the total distance covered by the piston during one cycle, τ is the
duration of one cycle, and δ is a fixed numerical parameter characterizing the
dependence of the velocity with respect to time (Rebhan 2002). As a result,
the power of the engine will be expressed by
P =
Q
1
−Q
2
τ

4fδD
2
τ
2
. (5.33)
For given values of Q
1
and Q
2
, P is a function of τ only. The maximum power
is then obtained from the condition that dP(τ)/dτ = 0, which yields for the
optimum time
τ
opt
=
8fδD
2
Q
1
−Q
2
. (5.34)
Introducing this result into (5.33), it is found that the maximum power is
P
max
=
(Q
1
−Q
2
)
2
16fδD
2
. (5.35)
The efficiency is given by
η =
W
Q
1
=
Q
1
−Q
2
Q
1

4fδD
2
τQ
1
. (5.36)
Substituting the value for the optimum time, one obtains
η(max P) =
1
2
_
1 −
Q
2
Q
1
_
=
η
Carnot
2
, (5.37)
which is of course lower than Carnot’s efficiency, but also lower than the
Curzon–Ahlborn’s result (5.26) derived for heat losses only.
122 5 Finite-Time Thermodynamics
Just like (5.26), expression (5.37) does not depend explicitly on the fric-
tion coefficient. Combined effects of heat losses and friction lead to rather
cumbersome mathematical relations, which are certainly interesting from a
practical perspective, but which do not add essential new ideas to the above
considerations.
5.2 Economical and Ecological Constraints
In practical situations, there will be a compromise between maximum effi-
ciency and maximum power, and the priorities will be settled by economical
and ecological considerations. Note that, from an economic point of view, the
efficiency of the engine should be written as
η
economic

W Pr(W)
Q
1
Pr(Q
1
)
, (5.38)
where Pr(W) is the price at which the unit of work is sold and Pr(Q
1
) is
the price at which the unit of heat which must be supplied to the engine is
bought. The economical efficiency will incorporate non-physical information,
as the prices of fuel and work (electrical work, for instance). As a matter of
fact, the actual economical efficiency is much more complicated than (5.38),
because it should also include, for instance, the investment for building the
plant and operating costs (as, for instance, the salaries of the workers, other
indirect costs as taxes, etc.). Anyway, the price of fuel plays generally a
crucial role: if fuel is expensive and the work produced by the engine may
only be sold at low price, then the situation with the highest efficiency, i.e.
taking most advantage of the fuel, will be the most suitable one. On the
contrary, if fuel is cheap and if the work may be sold at high price, the
situation of maximum power will be preferred from an economic point of view.
Furthermore, the higher the power output is, the faster the investments will
be recovered, whereas the higher the efficiency, the faster the fuel expenses
will be recovered.
Now, if ecological constraints are incorporated, the criteria may change.
For instance, the lowest is the efficiency, the least ecological will be the sta-
tion, because it will consume more fuel to produce the same amount of work
and therefore will produce more pollution. In short, it can be said that the
situation with the highest efficiency is the most ecological one, whereas, if
fuel is sufficiently cheap, the operation at maximum power output will be
the most interesting from the economical standpoint. In general, the situa-
tion is rather complex, and will depend on particular circumstances, as for
instances, on the rate of demand in energy.
Another example where a similar conflict between efficiency and power
arises is in exothermic chemical reactions. Indeed, it is known from equilib-
rium thermodynamics that the efficiency of the reaction (i.e. the number of
5.2 Economical and Ecological Constraints 123
moles which are produced for a given initial quantity of reactants) is increased
when temperature is lowered, according to the Le Chatelier’s principle dis-
cussed in Chap. 1. However, from a kinetic perspective, the lower the temper-
ature is, the lower will be the reaction rate. Thus, lowering temperature has
two opposite effects: it increases the efficiency but it diminishes the reaction
rate. Which temperature is optimal cannot be decided simply from thermo-
dynamic arguments: economical and ecological aspects must be included in
the analysis. From an economic standpoint, if the price of the reactant is low,
the most satisfactory approach is the fastest one; however, from the ecological
point of view, the faster the procedure is, the higher is the consumption of
reactant and the higher is the pollution produced by the factory.
Both power and efficiency must therefore be considered in the assessment
of actual heat engines. Up to now, we have discussed only two criteria, namely
power optimization and efficiency optimization. Other elements may be taken
into account to convey the subtleties of the interplay between economical and
ecological needs.
One proposal is to minimize the entropy production, which is related to
the loss of available work by W
lost
= T
2

irr
S; the rate of entropy production
over one cycle is, after use is made of expressions (5.12) for ∆
irr
S and (5.8)
and (5.9) for t
cycle
,
σ
s


irr
S
t
cycle
=
Q
1
_
1
T

1

1
T
1
_
+Q
2
_
1
T
2

1
T

2
_
Q
1
α(T
1
−T

1
)
+
Q
2
α(T

2
−T
2
)
. (5.39)
In terms of x
1
and x
2
defined under (5.16), this may be written as
σ
s
(x
1
, x
2
) =
α
1
α
2
T
1
T
2
x
2
1
x
2
T
2
+x
1
x
2
2
T
1
α
1
T
2
x
1

2
T
1
x
2
+ (α
1
−α
2
)x
1
x
2
. (5.40)
Another ecological criterion, which is often used, is the maximization of the
quantity
Ψ = power −T
2
entropy production = P −T
2

irr
S. (5.41)
By doing so, a high power (a factor of economical interest) is promoted,
together with low dissipation (a factor of ecological interest), in such a way
that one combines both economical and ecological aims. Physically, (5.41)
is related to the useful work available in a process, which is given by W =
W
R
− T
2

irr
S, W
R
being the work performed by the reversible process and
T∆
irr
S the work dissipated as heat by irreversibilities. Dividing both terms
of this expression by the cycle duration t
cycle
gives back (5.41). From the
expressions (5.17) for the power output and (5.39) for the rate of entropy
production, one obtains
Ψ(x
1
, x
2
) =
α
1
α
2
T
1
T
1
(T
1
−T
2
) −(T
1
+T
2
)x
1
−2T
1
x
2
α
1
T
2
x
1

2
T
1
x
2
+ (α
1
−α
2
)x
1
x
2
x
1
x
2
. (5.42)
124 5 Finite-Time Thermodynamics
The values of the variables for which this quantity becomes a maximum, in
the particular case α
1
= α
2
, are
x
1,max
=
T
1
2
T
1
−T
2
T
1
+T
2
+ 2T
1
A
, x
2,max
= Ax
1,max
, (5.43)
where
A ≡
_
T
2
(T
1
+T
2
)
2T
2
1
_
1/2
. (5.44)
Introducing (5.43) into (5.25), it is found that the corresponding efficiency is
η(at Ψ
max
) = η
Carnot
1 + 2
T
2
T
1
+
3
2

2
_
T
2
T
1
+
T
2
2
T
2
1
_
1/2
1 + 3
T
2
T
1
+ 2

2
_
T
2
T
1
+
T
2
2
T
2
1
_
1/2
. (5.45)
As an illustration, Table 5.1 gives the values of efficiency, power, entropy
production, and dissipation (product of lowest temperature times entropy
production) for a Carnot’s heat engine with T
1
= 400 K, T
2
= 300 K,
V
max
/V
min
= 2, α = 209 WK
−1
, under different operational conditions, as
considered by Angulo-Brown (1991). Here, it has been taken into account
that Q
1
= RT

1
ln(V
max
/V
min
) and Q
2
= RT

2
ln(V
max
/V
min
).
Note the practical advantage of the Ψ optimization criterion: though it re-
duces somewhat the power availability with respect to the maximum power
(from 375 to 294 W, i.e. a 25% reduction approximately), it lowers appre-
ciably the entropy production (from 1.08 to 0.3 WK
−1
) and the dissipation
(from 325 to 93 W). Therefore, it yields a satisfactory value for the power
output, which is economically interesting, and it considerably brings down
the dissipation with respect to the maximum power situation, an appreciated
result from the ecological point of view.
Table 5.1 Values of efficiency, power, entropy production, and dissipation for a
Carnot’s heat engine with T
1
= 400 K, T
2
= 300 K, V
max
/V
min
= 2, α = 209 WK
−1
,
under operation conditions defined by (a) maximum efficiency (Carnot), (b) maximum
power output, (c) minimum entropy production, and (d) maximum Ψ (from Angulo-
Brown, J. Appl. Phys. 69 (1991) 7465)
Maximum Maximum Minimum Maximum
efficiency power entropy production Ψ
Efficiency (%) 25 13 23 19.2
Power (W) 0 375 114 294
Entropy production (WK
−1
) 0 1.08 0.30 0.31
Dissipation (W) 0 325 90 93
5.3 Earth’s Atmosphere as a Non-Equilibrium System and a Heat Engine 125
5.3 Earth’s Atmosphere as a Non-Equilibrium System
and a Heat Engine
Planetary atmospheres are fascinating non-equilibrium thermodynamic sys-
tems, whose study has benefited by the impressive observational data pro-
vided by space missions. It is well recognized that Earth’s atmosphere is of
utmost importance for our survival: atmosphere warming and the climate
change are among the main topics in the current agenda of Earth’s sciences
and a topic of much social, economical, and political concern since the last
decade. The atmosphere is continuously subject to energy flow, because it ab-
sorbs, reflects, transforms, and stores shortwave solar radiation energy, and
releases back to the outer space longwave, infrared radiation energy. It repre-
sents therefore a paradigmatic example of a non-equilibrium system (Peixoto
and Oort 1984; Kleidon and Lorenz 2005). In this section, we present a sim-
plified version of the Earth energy balance, with special attention to the
understanding of global warming, and we apply finite-time thermodynam-
ics to evaluate the efficiency of the atmosphere as a heat engine, converting
thermal radiation energy into wind kinetic energy.
5.3.1 Earth’s Energy Balance
We present here an outline of the main factors influencing the average Earth
temperature, based on a rather simplified modelling of the global energy
balance. On the one hand, the Earth absorbs an incoming radiation power
P
abs
= (1 −r)πR
2
E
I
S
, (5.46)
where R
E
is the radius of the Earth and r is the reflectivity (also called
albedo), whose mean current value is 0.35. The quantity I
S
(≈1,370 Wm
−2
)
is the so-called solar constant, namely the maximum radiation flux arriving
at the upper part of the atmosphere, in a region which is directly facing
the Sun, and averaged over all latitudes and seasons and over the day/night
alternation. The combination πR
2
E
is thus the transversal area of the Earth,
intercepting the solar radiation. In virtue of Wien’s law of radiation, stating
that λ
max
=2,897T
−1
µm, with T the temperature of the body emitting
radiation, the radiation arriving from the Sun, whose surface temperature is
≈6,000 K, is shortwave. Except for reflection and absorption in the clouds,
the atmosphere is essentially transparent to this kind of radiation, which is
directly striking the surface of the Earth.
The power emitted by the Earth is
P
emit
= 4πR
2
E
eσT
4
E
, (5.47)
where e is the emissivity, whose approximate value is 0.58, σ the Stefan–
Boltzmann constant, and T
E
the average temperature of the Earth, a quan-
tity on which we will centre our attention. The numerical factor 4 in (5.47)
126 5 Finite-Time Thermodynamics
assumes that the Earth is radiating over the whole surface, and not only the
surface facing to the Sun. According to Wien’s law, the radiation emitted by
the Earth is longwave, and behaves as infrared radiation. As commented be-
low, several gases in the atmosphere absorb and re-emit this type of radiation,
in contrast to the incoming radiation.
To find the average temperature of the Earth, the emitted and absorbed
powers are made equal, i.e.
(1 −r)πR
2
E
I
S
= 4πR
2
E
eσT
4
E
, (5.48)
from which follows that
(1 −r)
I
S
4
= eσT
4
E
, (5.49)
and the resulting average temperature of Earth is given by
T
E
=
_
(1 −r)I
S
4eσ
_
1/4
, (5.50)
and found to be of the order of 290 K. In the balance equation (5.48), some
internal sources of heat – volcanoes and radioactive sources – have been
neglected, but should be incorporated in a more detailed analysis. According
to (5.50), the Earth reflectivity r and emissivity e have a deep influence on
the Earth temperature; for instance, a reduction of these parameters implies
an increase of the average temperature. If there was no greenhouse effect,
whose effect is to reduce e, the temperature of the Earth would be some
30

C colder than it is today.
The reflectivity, or albedo, r and the emissivity e are fractional quantities
with values comprised between 0 and 1; for a perfect black body, one would
have r = 0 and e = 1. The albedo r and emissivity e depend on several
variables, mainly temperature and composition. The actual value of r results
from a balance between the extension of ices, snows, deserts, and clouds,
which have high values of albedo (high reflectivity) and the surface occupied
by forests and seas, with low values.
5.3.2 Global Warming
Global warming is nowadays the object of an enormous amount of research
at the theoretical, observational, and numerical levels. The role of human
activities and their implications on climatic changes actually deserves an
increasing international interest. We will focus here on some qualitative ideas
which allow seizing the key features of these phenomena. Truly, the first
studies about global warming were carried out by Arrhenius in 1896, but
they did not receive a widespread attention until the end of the 1960s, when
the first data on the increase of CO
2
in the atmosphere became available.
5.3 Earth’s Atmosphere as a Non-Equilibrium System and a Heat Engine 127
The average value of Earth’s temperature is given by (5.50). Concerning
the emissivity e, its value depends especially on the concentration of those
gases sensible to infrared radiation. The major natural infrared absorbers in
the atmosphere are water vapour, clouds, and carbon dioxide. The current
concerns are related with CO
2
and some other gases, mainly CH
4
, N
2
O, and
CFC, whose concentration has considerably increased during the last cen-
tury as a consequence of human activities. For instance, the abundance of
CO
2
in the atmosphere has increased from 315 ppmv (parts per million in
volume) in 1950 to 350 ppmv in 1990, as a consequence of fossil fuel burning
and deforestation (for the sake of comparison, let us mention that in 1750
the CO
2
atmospheric concentration was some 280 ppmv). CO
2
remains in
the atmosphere for about 500 years. The increase in CH
4
has been much less
(1% per year, at the current rate). This gas is produced by bacteria that
decompose organic matter in oxygen-poor environments, as marshes or rice
paddles, burning of forests and grasslands, or the guts of termites in the di-
gestive tracts of cattle, sheep, pigs, and other livestock. CH
4
remains in the
atmosphere for 7–10 years and each molecule is some 25 times more efficient
than a molecule of carbon dioxide in warming the atmosphere. Other gases
with much effect are CFC (chlorofluorocarbons), which also deplete ozone
in the stratosphere. The main sources are leaking air conditioners and re-
frigerators, or evaporation of some industrial solvents. CFC remain in the
atmosphere for 65–110 years, and a molecule of CFC has 10,000–20,000 times
the impact on global warming compared to a molecule of carbon dioxide.
Nitrous oxide (N
2
O) is released mainly from the breakdown of nitrogen fer-
tilizers in soil, and nitrate-contaminated groundwater. Its average stay in the
atmosphere is 150 years and the effect on global warming from each molecule
is 230 times more important than a CO
2
molecule. The main contributions
to the greenhouse effect are 54% CO
2
, 18% CH
4
, 14% CFC, and 6% N
2
O.
The net planetary radiative forcing is dominated by the presence of the
water vapour whose contribution to the total value of e is of the order of
60%, and whose concentration is strongly related to that of CO
2
. Note that a
relatively small variation of temperature may produce an important feedback
in the water vapour concentration. Another possible multiplicative feedback
arises from the relationship between ice extension and the albedo r: a warmer
planet will have less ice meaning a reduction in the reflectivity, which in
turn will produce an increase of the temperature, thus leading to a further
reduction of ice extension and so on.
The main uncertain factor influencing global warming and its impact on
climate concerns the role of clouds on the radiative balance because clouds
contribute both to warming, by absorbing infrared radiation, and to cooling,
by reflecting a part of the incoming radiation. The relative importance of
these two antagonistic contributions depends on the altitude of the clouds:
for low ones, warming plays the leading part, whereas for high ones, cooling is
the dominating effect. Another source of query is the response of the ocean to
variations of the concentration of the greenhouse gases and to perturbations
128 5 Finite-Time Thermodynamics
in the oceanic currents distributing the heat from the Equator to the Poles.
Furthermore, since the oceans have an enormous thermal inertia, the warming
process is expected to be slow; this means that it may take many years before
warming effects become perceptible.
The estimations about the CO
2
atmospheric concentration are deeply
bound to the evolution of CO
2
emissions. It is estimated that, in 2100, CO
2
concentration could be close to 550 ppmv. Solutions are not easy to find be-
cause of the clash of interests between the necessity to reduce the use of fossil
fuels and the increasingly widespread and growing need for energy, thus lead-
ing to serious policy confrontations.
Useful information about climate change may also be drawn from a study
of the global entropy balance of Earth. The convective motions of upward hot
air and down falling cold air and several phenomena observed in our everyday
life imply the appearance of some structure and order, which must be com-
patible with a global increase of entropy of the universe. It turns out that,
globally, the Earth is a very dissipative system. Indeed, it receives photons
at T
S
≈ 6,000 K and emits them at T
E
≈ 290 K. For each incoming photon,
some 20 photons are re-emitted to the outer space. Other sources of entropy
production, as for instance heat transfer between hot and cold regions, or
turbulence of winds and oceanic currents are unavoidable and should be in-
corporated in more detailed analyses. Several models taking into account the
role of entropy production on the climate have been proposed. Based on sim-
plified atmospheric models, Nicolis and Nicolis (1980) and Paltridge (1981)
explored whether or not the climate is governed by an extremal principle re-
lated with the entropy production. They found reasonable values for the most
significant climatic variables by maximizing the transport part of the entropy
production. We will not enter into this rather specialized topic, which throws
anyway a light on the contribution of the second law to a better understand-
ing of the global climate system. For more recent models, see for instance
Pujol and Llebot (1999) and Ozawa et al. (2003).
5.3.3 Transformation of Solar Heat into Wind Motion
In this section, we will apply finite-time thermodynamics to the Earth at-
mosphere, considered as a heat engine driven by solar radiation and pro-
ducing useful work in the form of wind motion. On the yearly average, the
quantity of heat transferred per unit time by solar radiation to the surface
of the Earth is, per unit area,
˙
Q
S
=
1 −r
4
I
S
≈ 223 Wm
−2
, (5.51)
where, as in Sect. 5.3.2, I
S
is the solar constant and r is the albedo. Since the
average energy of wind motion is approximately 7 Wm
−2
, the efficiency of
5.3 Earth’s Atmosphere as a Non-Equilibrium System and a Heat Engine 129
conversion of solar radiation energy into mechanical energy of wind is about
7/223 ≈ 3%.
A rough upper bound estimation for the conversion of solar heat into
work is obtained by taking for the temperature of the hot reservoir T
1
=
313 K, corresponding to the average Equator temperature of 40

C, and for
the temperature of the cold reservoir T
2
= 233 K, the average temperature
at the Poles. After introducing these values in the expression of the Carnot’s
efficiency, one finds 25.5%, which is much higher than the 3% observed value.
Applying the maximum power efficiency of the Curzon–Ahlborn’s engine, the
result falls down to 14%. However, instead of Curzon–Ahlborn’s approach,
we prefer to follow an analysis by Gordon and Zarmi (1989), which is more
illustrative and pedagogical, by keeping nevertheless simplicity.
These authors investigate that the global-scale motion of wind in convec-
tive Hadley cells, namely those cells in which hot air rises towards colder
regions along the upper atmosphere, falls down on the surface of the Earth,
and moves back to hotter regions along the Earth surface. Gordon and Zarmi
model these various motions by means of a Carnot’s engine, with two isother-
mal and two adiabatic branches. The first isothermal corresponds to the at-
mosphere absorbing solar energy at low altitudes, at temperature T

1
, not
exactly equal to the Earth surface temperature T
1
. The second isothermal
branch, at temperature T

2
, is related to the atmosphere at high altitudes,
where air rejects heat to the universe. The adiabatic branches correspond to
the rising and falling air currents. Furthermore, it is assumed that the time
spent on the isothermal branches is equal, and that the time spent on the
adiabatic branches is negligible. The values of the hotter T

1
and colder tem-
peratures T

2
of the working fluid will be determined to maximize the power
conversion, in analogy with Sect. 5.1.
In the present case, the heat exchange mechanisms are of radiative type,
and therefore the analysis of Sect. 5.1, based on Newton’s law of transfer, is
not directly applicable. This justifies that we re-examine the problem. The
power (per unit area) available for work will be
P =
˙
Q
1

˙
Q
2
=
˙
Q
S
−σ
_
T
4
1
2
+
T
4
2
2
_
, (5.52)
with the identifications
˙
Q
1
=
˙
Q
S

1
2
σT
4
1
,
˙
Q
2
=
1
2
σT
4
2
, and σ = 5.67
10
−8
Wm
−2
K
−4
the Stefan–Boltzmann constant, an upper dot means time
derivation. The factor
1
2
in (5.52) is a consequence of the assumption that
air spends half the time on one isotherm and half the time on the other. We
know from Sect. 5.1 that this does not exactly reflect the reality, but since we
are only interested in obtaining some orders of magnitude, it is not extremely
important.
On the other side, the entropy balance may be expressed as

˙
Q
1
T

1
+
˙
Q
2
T

2
= −
˙
Q
S
T

1
+
σT
3
1
2
+
σT
3
2
2
= 0, (5.53)
130 5 Finite-Time Thermodynamics
where it is assumed that the cycle is endoreversible and the entropy flux is
simply
˙
Q
i
/T, with
˙
Q
i
given by the same relations as in (5.52).
The temperatures T

1
and T

2
are derived by maximizing the mechanical
power (5.52), subject to the constraint (5.53). Since
˙
Q
S
is fixed, we may
maximize
Φ(T

1
, T

2
) ≡
σ
2
_
T
4
1
+T
4
2
_

_
˙
Q
S
T

1

σT
3
1
2

σT
3
2
2
_
, (5.54)
where λ is a Lagrange multiplier taking into account condition (5.53). The
final results are
T

1,max
= 229 K, T

2,max
= 192 K, and P
max
= 17.1 Wm
−2
. (5.55)
The resulting efficiency is 7%, much closer to the actual one (3%) than the
Carnot’s and Curzon–Ahlborn’s na¨ıve estimations.
In more sophisticated formulations, the hot isotherm is considered to be
on the dayside of the Earth and the cold one on the nightside. Dissipation
of the wind energy is also introduced in some analyses. Furthermore, instead
of assuming a homogeneous model for the Earth, one may consider several
convective cells (the so-called Hadley cells) from the Equator to the Pole.
Including all these refinements yields results that are very close to the actual
observed values.
5.4 Problems
5.1. Heat engine with heat leak. Assume that, parallel to the Curzon–Ahlborn’s
engine studied in Sect. 5.1, there is a direct heat leak from the hot source at
temperature T
1
and the cold source at temperature T
2
. Then the efficiency
must be defined as
η(τ) ≡
W(τ)
Q
leak
(τ) +Q
1
(τ)
=
W(τ)
Q
leak
(τ) +
W(τ)
1−τ
,
where τ ≡ T

2
/T

1
and Q
leak
(τ) = K
leak
t
total
(τ)(T
1
−T
2
), with K
leak
the heat
transfer coefficient of the leak and t
tot
(τ) the total duration of the cycle for
the Curzon–Ahlborn’s engine. Note that, in contrast with the latter, Carnot’s
efficiency is never reached, because due to the heat leak, Q
leak
diverges, and
the efficiency of this very slow cycle is very low due to the high value of the
heat leak. Find the maximum efficiency and the maximum power for this
engine.
5.2. The Curzon–Ahlborn’s engine. Show that, in the Curzon–Ahlborn’s en-
gine, the ratio t
1
/t
2
associated to the two isotherm branches is given by
t
1
/t
2
= (α
2

1
)
1/2
.
5.4 Problems 131
5.3. Maximizing Curzon–Ahlborn’s engine. Assume that the total duration
of the cycle given by (5.8) and (5.9) is fixed. Then, maximizing the power is
equivalent to maximize the difference Q
1
−Q
2
. Performing this maximization,
by introducing the constraint on the duration of the cycle by means of a
Lagrange multiplier λ, i.e. by imposing
δ
_
Q
1
−Q
2

_
Q
1
α
1
(T
1
−T

1
)
+
Q
2
α
2
(T

2
−T
2
)
__
= 0
and recalling (5.13), show that
x
1,max
x
2,max
=
_
α
2
T
1
α
1
T
2
_
1/2
,
which is the result (5.21).
5.4. Minimum entropy production criterion. Show that the minimization of
the total entropy production ∆
irr
S, under the condition of fixed total duration
of the cycle, given by (5.8) and (5.9), yields
x
1,max
x
2,max
=
_
α
2
α
1
_
1/2
T
1
T
2
.
Note that this result is different from the previous problem.
5.5. Maximization process in Curzon–Ahlborn’s engine. Assume that the en-
vironment temperature is T
0
, in such a way that the loss power is T
0
σ instead
of T
2
σ as taken in (5.41). By analogy with (5.41), study the maximization of
Ψ
0
≡ P − T
0
σ, with T
0
constant, and show that the efficiency at maximum
Ψ
0
is (Z. Yan, J. Appl. Phys. 73 (1993) 3583)
η(at max Ψ
0
) = 1 −
T
2
T
1
_
T
1
+T
0
T
2
+T
0
_
1/2
.
5.6. Finite-time analysis of refrigerators. One of the basic quantities in re-
frigerators is the so-called coefficient of performance (COP), defined as
COP ≡
heat extracted from the cold source
work supplied to the engine
.
Performing an analysis similar to that of the Curzon–Ahlborn’s heat engine,
find the coefficient of performance at maximum cooling power. Hint: see
Agrawal and Menon 1990.
5.7. Maximum power of solar thermal heat engines. A simple endoreversible
modelling of solar thermal heat engines is to assume that the heat transfer
between the hot source and the engine is given by a Stefan–Boltzmann law
132 5 Finite-Time Thermodynamics
˙
Q
1
= K(T
4
1
−T
4
1
), and that the heat transfer to the cold source is reversible.
Prove that the power–efficiency relation is
P(η) =
ηK
_
(1 −η)
4
T
4
1
−T
4
2
¸
(1 −η)
4
,
with η ≡ 1 − (T
2
/T

1
). Show that 4T
5
1
− 3T
2
T
4
1
− T
4
1
T
2
= 0 corresponds to
the maximum power.
5.8. Energy conversion efficiency of a simple photovoltaic cell. Assume that
the device is characterized by only two energy levels with a distance apart
equal to E
G
. Photons, with energy less than E
G
, simply move through the
device without leaving to it any energy. Photons with energy higher than
E
G
excite particles from the lower to the higher levels, and perform a useful
work equal to E
G
whereas the excess energy is dissipated as heat. The energy
efficiency may be defined as
η ≡
useful power
incident power
=
E
G
_

E
G
E
2
dE
exp(βE)−1
_

E
G
E
3
dE
exp(βE)−1
,
where β = (k
B
T)
−1
. In the second equality, it is assumed that the energy
distribution of the incoming radiation is Planck’s black body distribution
at temperature T. (a) Show that the maximum efficiency is achieved for
materials whose energy gap E
G
is E
G,opt
= 2.17k
B
T. (b) Estimate E
G
if
T =5,760 K, the Sun surface temperature. (c) Evaluate the maximum effi-
ciency. Truly the actual efficiency will be less than this value (see, for instance,
Landsberg and Tonge 1980; de Vos 1992).
5.9. Average temperature of the atmosphere. Determine the value of the av-
erage Earth temperature from (5.50) in the following cases. (a) First, assume
that the reflectivity r is zero and the emissivity e is unit, corresponding to the
case that the Earth behaves as a perfectly black body. (b) The reflectivity r
has the actual value, 35%, and e = 1 (no greenhouse effect), and compare the
result with the current average temperature, namely 288 K. (c) Determine
the value of the emissivity e in order that Earth’s average temperature has
its current actual value.
5.10. Na¨ıve approach to global warming. A main feature of global warm-
ing due to increase of atmospheric CO
2
concentration is that emissivity e is
lowered, because CO
2
absorbs the infrared radiation emitted by the Earth.
Assume, for simplicity, that e = 0.58 − 0.53 ln (c
CO
2
/c
CO
2
,0
), where c
CO
2
is
the CO
2
concentration in the atmosphere and c
CO
2
,0
≈ 350 ppmv is a refer-
ence value (in parts per million in volume). Determine the temperature in-
crease if the CO
2
concentration is raised, respectively, to 700 and 1,400 ppmv.
Comment: This model is pedagogically illustrative but excessively simplistic,
5.4 Problems 133
because it ignores important feedbacks of the temperature increase on the
water vapour content, the cloud formation, and the albedo.
5.11. Sun’s and Earth’s temperatures. The current value of the solar con-
stant, namely the maximum radiation flux arriving at the upper part of
the atmosphere in a region, which is directly facing the Sun, is equal to
1,367 Wm
−2
. (a) Calculate the temperature at the surface of the Sun, by
assuming that it radiates as a black body. (b) Determine the variation of
the solar constant if the temperature of the Sun is increased by 1% with
respect to its current value, obtained before. (c) To which extent will the
average temperature of the Earth change if the albedo and the emissivity
were kept constant? (Stefan–Boltzmann constant, σ = 5.67 10
−8
Wm
−2
;
distance Sun–Earth = 1.50 10
11
m, radius of the Sun = 6.96 10
8
m).
5.12. Carnot engine at finite time. A Carnot engine is working between two
heat reservoirs at 400 K and 300 K. a) Compare the maximum efficiency with
the efficiency at maximum power. b) If the engine receives 1000 J from the
hot reservoir, evaluate the entropy produced per cycle in the situation at
maximum power. c) If the duration of the cycle at maximum power is 20 s,
find the value of the maximum power delivered by the engine. d) If the
transfer coefficients between the engine and the reservoirs do not depend
on temperature, find the efficiency and power of this engine when the cycle
duration is 60 seconds.
5.13. Ice comet. A spherical ice comet of 10
5
Kg describes an eccentric ellip-
tical orbit around the Sun, with major radius 2 10
8
km and minor radius
1 10
8
km. Evaluate which proportion of the mass of the comet will evaporate
during one orbit. Assume that the albedo coefficient is 0.8 for ice and 0.2 for
liquid water.
5.14. Na¨ıve scaling relation for the life-span of stars. Bigger stars have
shorter lifespans than smaller stars. The physical idea is that they need
higher internal pressures, and therefore higher temperatures, to resist the
higher gravitational pressures. Since the energy emitted per unit time, which
is proportional to the loss of mass of the star per unit time in nuclear re-
actions, is proportional to T
4
, according to Stephan-Boltzmann’s law, they
will consume faster their mass. a) Make a simple estimation of the scaling
exponent α relating the lifespan of the star t
star
to the mass M of the star
through t
star
= AM
−α
, with A a constant. (Hint: equate the pressure of an
ideal gas to the average gravitational pressure determined on dimensional
grounds from G, M and R, the gravitational constant and the mass and
radius of the star. Note that if one makes the simplistic assumption that
different stars have the same density, it is found that t
star
= AM
−7/3
). b)
According to this estimate, how long will last a star of twice the mass of the
Sun, if the total lifespan of the Sun will be of some 10
10
years? Will such a
star be able to sustain intelligent life in its planetary system, if any?
134 5 Finite-Time Thermodynamics
5.15. Minimum entropy production. a) Find the hot and cold temperature
ratios x
1
and x
2
defined in (5.16) which extremize the entropy production
(5.40) (assume, for simplicity, that α
1
= α
2
). b) Determine the corresponding
value for the efficiency as obtained from (5.25). c) Evaluate the correspond-
ing efficiency and compare it with Carnot’s maximum efficiency and with
efficiency at maximum power.
Chapter 6
Instabilities and Pattern Formation
Dissipative Structures Far from Equilibrium
The notion of irreversibility is frequently associated with a tendency of the
systems towards spatial homogeneity and rest, i.e. towards equilibrium. Al-
though this observation is true in isolated systems, it is no longer verified in
closed and open systems, which may exhibit organized structures in space and
time. This organization arises as a consequence of occurrence of instabilities
in systems driven far from equilibrium by external forces. Instabilities may
be considered as phase transition phenomena, because they manifest under
the form of a discontinuous change of the physical behaviour at a critical
value of an external control parameter.
Such non-equilibrium processes are, however, rather different from those
occurring in equilibrium phase transitions. In equilibrium thermodynamics,
there is an evident link between the second law and stability. In Chap. 1,
we have derived an explicit criterion of stability by imposing that the sec-
ond variation of the entropy is negative definite. In non-equilibrium, such a
universal criterion of stability does not exist because it is generally not pos-
sible to construct thermodynamic potentials depending on the whole set of
variables. The general problem of stability is then conventionally carried out
by submitting the reference state, for instance, a stationary non-equilibrium
state, to an arbitrary disturbance. These disturbances may be caused delib-
erately from the outside by external agents, like a temperature or pressure
gradient, or induced by irregularities or imperfections of the apparatus or
still by internal fluctuations of the system (see Chap. 11).
A proof of stability is based on the time variation of the perturbation:
if the perturbation decays in a finite time, the reference state is stable, and
it is asymptotically stable if the disturbance tends to zero in the long time
limit. If the disturbance grows with time, the reference state is unstable. In
this case, the mathematical solution of the problem cannot survive in the
physical world and the behaviour of the system will be governed by another
mathematical solution: it is said that the solution suffers a bifurcation. When
the disturbance does neither decrease nor increase with time, one speaks
135
136 6 Instabilities and Pattern Formation
about neutral or marginal stability, and the corresponding values taken by
the relevant parameters are said to be critical.
When the perturbations remain infinitesimally small, the approach is lin-
ear. Several illustrations – like B´enard–Marangoni’s convective instability,
Taylor’s instability, and Turing’s reaction–diffusion instability – are exam-
ined in the present chapter. One of the most interesting aspects of the sta-
bility theory is the prediction of organized structures, taking the form of
spatial and temporal patterns, which appear beyond the linear critical insta-
bility threshold. These patterns are also referred to as dissipative structures,
a term introduced by Prigogine to characterize ordered systems maintained
far from equilibrium by external constraints. The mechanisms underlying the
development of patterns are essentially non-linear and modelled by coupled
non-linear partial differential equations whose treatment is rather intricate.
Various examples are treated in the second part of this chapter.
In Sects. 6.1 and 6.2, we briefly discuss the mathematical aspects of the
linear and non-linear theories of stability. When the perturbations are infin-
itesimally small, the problem is linear and the solution may be decomposed
into normal modes. This technique has been made popular by the celebrated
textbook of Chandrasekhar (1961). As soon as the amplitude of the pertur-
bations becomes finite, the linear approach is no longer appropriate and must
be replaced by more sophisticated non-linear techniques or direct numerical
solutions. A detailed analysis of these methods can be found in specialized
monographs (Joseph 1976; Drazin and Reid 1981; Cross and Hohenberg 1993;
Bodenshatz et al. 2000); here we shall essentially focus on the so-called am-
plitude method which, in our opinion, is the simplest and amongst the most
powerful.
One of the most attractive aspects of the (non-linear) stability theory is the
prediction of self-organization phenomena, in the form of spatial or tempo-
ral patterns, beyond the critical threshold. Pattern formations are important
and appear in a multitude of natural phenomena being a central theme of
research in modern physics. They provide a major feature of the dynamics of
several processes observed not only in hydrodynamics but also in electricity,
optics, material science, oceanography, geophysics, astrophysics, chemistry,
and living systems. Thermal convection in fluids is one of the most repre-
sentative problems and is discussed in Sect. 6.3. Indeed, most of the recent
developments about non-equilibrium processes in the context of instabilities
and chaos have been widely inspired from the study of this example.
In this chapter, the approach is more dynamical than thermodynamical,
i.e. it is based on the study of evolution equations rather than on state and
constitutive thermodynamic relations. However, from a conceptual point of
view, it is very important to understand that thermodynamics does not pre-
clude the formation of patterns provided the systems are driven far enough
from equilibrium. Such a topic represents an important contribution towards
a better knowledge of non-equilibrium systems and therefore it finds naturally
a place in this book.
6.1 The Linear Theory of Stability 137
6.1 The Linear Theory of Stability
This theory is concerned with stability of systems subject to infinitesimally
small perturbations about a reference state. Most physical and chemical sys-
tems obey evolution equations of the form
da
dt
= f(a, ∇a; r, t), (6.1)
where f is a continuous and twice differentiable function depending on the
field variables a, for instance the velocity v and the temperature T fields, and
their gradients, r designates the position vector, and t the time. Assume that
there exists a “reference” solution corresponding to some specific constraints
a = a
r
. (6.2)
Such a solution is also called unperturbed or basic, and may be time de-
pendent (for instance periodic) or time independent (i.e. stationary). The
question now arises about the stability of this particular solution. Therefore,
one examines the response of the system with respect to a disturbance a

(r, t)
so that the generalized coordinates a(r, t) become
a(r, t) = a
r
+a

(r, t), (6.3)
where the quantity a

(r, t) is the “perturbation” which is supposed to be
small enough so that all non-linear terms can be neglected in (6.1). Substi-
tuting (6.3) in the evolution equations (6.1) and developing them around the
reference solution yields a linear set of differential equations of the form
da

dt
= La

, (6.4)
where the linear operator L is to be evaluated at the reference unperturbed
state. Since linear stability implies stability with respect to all possible infin-
itesimal disturbances, we have to examine the reaction of the system to all
such disturbances. Practically, this will be accomplished by expressing a

as
a superposition of a complete ensemble of n normal modes, generally selected
as Fourier modes in a two-dimensional wave vector space (k = k
x
, k
y
),
a

(x, z, t) =

n
W
n
(z) exp(σ
n
t) exp(ik
n
x), (6.5)
and by examining the stability with respect to each individual mode n; x(x, y)
is the position vector in the plane normal to the z-axis. The quantity W
n
(z)
is called the amplitude and σ
n
is the growth rate of the disturbance,
σ
n
= Re σ
n
+ i Imσ
n
, (6.6)
138 6 Instabilities and Pattern Formation
generally a complex and k-dependent quantity. The k dependence of the
growth rate describes the spatial symmetry of the system; in rotationally
invariant systems, the σ
n
’s will only depend on the modulus [k[, whereas in
anisotropic systems like nematic liquid crystals there is an angle between the
direction of anisotropy and k.
The requirement that the field equations have non-trivial solutions leads
to an eigenvalue problem for the σ
n
’s. The stability problem is completely
determined by the sign of the real part of the σ
n
’s:
• If one single Re σ
n
> 0, the system is unstable.
• If Re σ
n
< 0 for all the values of n, the system is stable.
• If Re σ
n
= 0, the stability is marginal or neutral.
In the case of marginal stability, there corresponds to each value of k a critical
value of the control parameter, say the temperature difference ∆T in B´enard’s
problem, a characteristic velocity in flows through a pipe or the angular
velocity in Taylor’s problem, for which Re σ
n
= 0. All these critical values
define a curve of marginal stability, say ∆T vs. k, whose minimum (∆T
c
, k
c
)
determines the critical threshold of instability. In stability problems, it is
convenient to work with non-dimensional control parameters like the Rayleigh
number Ra in B´enard’s instability, the Reynolds number Re for the transition
from laminar to turbulent flows, or the Taylor number Ta in presence of
rotation; therefore, the marginal curves and the corresponding critical values
are generally expressed in terms of these non-dimensional quantities.
In several problems, it is postulated that Re σ = 0 implies Imσ = 0, this
conjecture is called the principle of exchange of stability, which has been
demonstrated to be satisfied in the case of self-adjoint problems; in this case
a stationary state is attained after the onset of the instability. If Re σ = 0
but Imσ ,= 0, the onset of instability is initiated by oscillatory perturbations
and one speaks of overstability or Hopf bifurcation. This kind of instability
is observed, for instance, in rotating fluids or fluid layers with a deformable
interface. The condition Re σ
n
> 0 for at least one value of n is a sufficient
condition of instability; on the contrary, even when all the eigenvalues are
such that Re σ
n
< 0, one cannot conclude in favour of stability as one can-
not exclude the possibility that the system is unstable with respect to finite
amplitude disturbances. It is therefore worth to stress that a linear stability
analysis predicts only sufficient conditions of instability.
6.2 Non-Linear Approaches
As soon as the amplitude of the disturbance is finite, the linear approach is
not appropriate and must be replaced by non-linear theories. Among them
one may distinguish the “local” and the “global” ones. In the latter, the de-
tails of the motion and the geometry of the flow are omitted, instead attention
6.2 Non-Linear Approaches 139
is focused on the behaviour of global quantities, generally chosen as a positive
definite functional. A typical example is Lyapounov’s function; according to
Lyapounov’s theory, the system is stable if there exists a functional Z satisfy-
ing Z > 0 and dZ/dt ≤ 0. In classical mechanics, an example of Lyapounov’s
function is the Hamiltonian of conservative systems. Glansdorff and Prigogine
(1971) showed that the second variation of entropy δ
2
S provides an exam-
ple of Lyapounov’s functional in non-equilibrium thermodynamics. The main
problems with Lyapounov’s theory are:
1. The difficulty to assign a physical meaning to the Lyapounov’s functional
2. The fact that a given situation can be described by different functionals
3. That in practice, it yields only sufficient conditions of stability
We do no longer discuss this approach and invite the interested reader to
consult specialized works (e.g. Movchan 1959; Pritchard 1968; Glansdorff
and Prigogine 1971). Here we prefer to concentrate on the more standard
“local” methods where it is assumed that the perturbation acts at any point
in space and at each instant of time. We have seen that the solution of the
linearized problem takes the form exp(σ
n
t) and that instability occurs when
the growth rate becomes positive, or equivalently stated, when the dimension-
less control parameter R exceeds its critical value R
c
. For values of R > R
c
,
the hypothesis of small amplitudes is no longer valid as non-linear terms be-
come important and will modify the exponential growth of the disturbances.
Another reason for taking non-linear terms into account is that the linear ap-
proach predicts that a whole spectrum of horizontal wave numbers become
unstable. This is in contradiction with experimental observations, which show
a tendency towards simple cellular patterns indicating that only one single
wave number, or a small band of wave numbers, is unstable.
Non-linear methods are therefore justified to interpret the mechanisms oc-
curring above the critical threshold. The problem that is set up is a non-linear
eigenvalue problem. Unfortunately, no general method for solving non-linear
differential equations in closed form has been presented and this has moti-
vated the development of perturbation techniques. A widely used approach
is the so-called amplitude method initiated by Landau (1965) and developed
by Segel (1966), Stuart (1958), Swift and Hohenberg (1977), and many oth-
ers. It is essentially assumed that the non-linear disturbances have the same
form as the solution of the linear problem with an unknown time-dependent
amplitude. Explicitly, the solutions will be expressed in terms of the eigen-
vectors W(z) of the linear problem in the form
a

(x, z, t) = A(t) exp(ik x)W(z), (6.7)
where A(t) denotes an unknown amplitude, generally a complex quantity.
In the linear approximation, A(t) is proportional to exp(σt) and obeys the
linear differential equation
dA(t)
dt
= σA(t), (6.8)
140 6 Instabilities and Pattern Formation
whereas in the non-linear regime, projection of (6.8) on the space of the W’s
leads to a coupled system of non-linear ordinary differential equations for the
amplitudes
dA(t)
dt
= σA(t) +N(A, A), (6.9)
where N(A, A) designates the non-linear contributions. Practically, the equa-
tions are truncated at the second or third order. A simple example is provided
by the following Landau relation (Drazin and Reid 1981)
dA
dt
= σA−l
c
[A[
2
A, (6.10)
where l
c
is a complex constant depending on the system to be studied and
[A[
2
= AA

with A

the complex conjugate of A. In (6.10), one has imposed
the constraint A = −A reflecting the inversion symmetry of the field variables
like the velocity and temperature fields. This invariance property is destroyed
and additional quadratic terms in A
2
will be present when some material
parameters like viscosity or surface tension are temperature dependent. To
take into account some spatial effects like the presence of lateral boundaries,
it may be necessary to complete the above relation (6.9) by spatial terms in
A or independent terms.
By multiplying (6.10) by A

and adding the complex conjugate equation,
one arrives at
d [A[
dt
2
= 2(Re σ) [A[
2
−2l [A[
4
, (6.11)
where l is the real part of l
c
. If A
0
designates the initial value, the solution
of (6.11) is
[A[
2
=
A
2
0
l
Re σ
A
2
0
+
_
1 −
l
Re σ
A
2
0
_
exp[−2(Re σ)t]
. (6.12)
(1) Let us first examine what happens for l > 0. When (Re σ) < 0, the system
relaxes towards the reference state A = 0 which is therefore stable; in
contrast, for (Re σ) > 0, the solution (6.12) tends, for t → ∞, to a
stationary solution [A
s
[ given by
[A
s
[ = (Re σ/l)
1/2
, (6.13)
which is independent of the initial value A
0
. This is a supercritical sta-
bility, the reference flow becomes linearly unstable at the critical point
Re σ = 0, or equivalently at R = R
c
, and bifurcates on a new steady
stable branch with an amplitude tending to A
s
. When the bifurcation is
supercritical, the transition between the successive solutions is continu-
ous and is called a pitchfork bifurcation as exhibited by Fig. 6.1.
It is instructive to develop Re σ around the critical point in terms of the
wave number k and the dimensionless characteristic number R so that
6.2 Non-Linear Approaches 141
Fig. 6.1 Supercritical pitchfork bifurcation: the solution A = 0 is linearly stable for
R < R
c
but linearly unstable for R > R
c
, the branching of the curve at the critical
point R = R
c
is called a bifurcation. Unstable states are represented by dashed lines
and stable states are represented by solid lines
Re σ = α(R −R
c
) +β(k −k
c
) (k −k
c
) + , (6.14)
where α is some positive constant. When R < R
c
, all perturbations
are stable with Re σ < 0; at R = R
c
, the system is marginally stable
and when R increases above R
c
, the system becomes linearly unstable.
Combining (6.14) with (6.13) results in
A
s
∼ (R −R
c
)
1/2
as R →R
c
, (6.15)
indicating that the amplitude A
s
of the steady solution is proportional to
the square root of the distance from the critical point. There is a strong
analogy with a phase transition of second order where the amplitude
A
c
of the critical mode plays the role of the order parameter and the
exponent 1/2 in (6.15) is the critical exponent.
(2) Let us now examine the case l < 0. If Re σ > 0, both terms of Lan-
dau’s equation (6.11) are positive and [A[ grows exponentially; it follows
from (6.12) that [A[ is infinite after a finite time t = (2Re σ)
−1
ln[1 −
(Re σ)/(lA
2
0
)], however this situation never occurs in practice because in
this case it is necessary to include higher-order terms in [A[
6
, [A[
8
, . . .
in Landau’s equation and generally no truncation is allowed. A more
realistic situation corresponds to Re σ < 0; now, the two terms in the
right-hand side of (6.12) are of opposite sign. Depending on whether
A
0
is smaller or larger than [A
s
[ given by (6.12), we distinguish two
different behaviours; for A
0
< [A
s
[, the solution given in (6.12) shows
that [A[ ≈ exp[(2Re σ)t] and tends to zero as t → ∞; in contrast
for A
0
> [A
s
[, the denominator of (6.12) becomes infinite after a time
t = (2Re σ)
−1
ln[1 − Re σ/lA
2
0
] and [A[ → ∞ (see Fig. 6.2). In this case,
142 6 Instabilities and Pattern Formation
Fig. 6.2 Time dependence of the amplitude for two different initial values A
0
in
the case of a subcritical instability
the reference state is stable with respect to infinitesimally small distur-
bances but unstable for perturbations with amplitude greater than the
critical value A
s
, which appears as a threshold value. This situation is
referred to as subcritical or metastable, by using the vocabulary of the
physicists.
In some systems, as for instance non-Boussinesq fluids, where the transport
coefficients like the viscosity or the surface tension are temperature depen-
dent, the symmetry A = −A is destroyed and the amplitude equation takes
the form
dA
dt
= σA+gA
2
−lA
3
, (6.16)
when A is assumed to be real, g and l are positive constants characterizing
the system. This form admits three steady solutions A
s
= 0 and A
1,2
given by
A
1,2
=
g ±
_
g
2
+ 4σl
2l
, (6.17)
and they are represented in Fig. 6.3 wherein the amplitude A
s
is sketched as
a function of the dimensionless number R.
For R < R
G
, the basic flow is globally stable which means that all pertur-
bations, even large, decay ultimately; for R
G
< R < R
c
, the system admits
two stable steady solutions A
s
= 0 and the branch GD whereas CG is un-
stable. At R = R
c
, the system becomes unstable for small perturbations and
we are faced with two possibilities: either there is a continuous transition
towards the branch CF which is called a transcritical bifurcation, character-
ized by the intersection of two bifurcation curves, or there is an abrupt jump
to the stable curve DE, the basic solution “snaps” through the bifurcation
to some flow with a larger amplitude. By still increasing R, the amplitude
will continue to grow until a new bifurcation point is met. If, instead, the
6.3 Thermal Convection 143
Fig. 6.3 Subcritical instability: the system is stable for infinitesimally small per-
turbations but unstable for perturbations with amplitude larger than some critical
value. Solid and dot lines refer to stable and unstable solutions, respectively
amplitude is gradually decreased, one moves back along the branch EDG up
to the point G where the system falls down on the basic state A = 0 iden-
tified by the point H. The cycle CDGH is called a hysteresis process and is
reminiscent of phase transitions of the first order.
In the present survey, it was assumed that the amplitude equation was
truncated at order 3. In presence of strong non-linearity, i.e. far from the
linear threshold, such an approximation is no longer justified and the intro-
duction of higher-order terms is necessary, however this would result in rather
intricate and lengthy calculations. This is the reason why model equations,
like the Swift–Hohenberg equations (1977) or generalizations of them (Cross
and Hohenberg 1993; Bodenshatz et al. 2000), have been recently proposed.
Although such model equations cannot be derived directly from the usual
balance equations of mass, momentum, and energy, they capture most of the
essential of the physical behaviour and have become the subject of very in-
tense investigations. Recent improvements in the performances of numerical
analysis have fostered the resolution of stability problems by direct integra-
tion of the governing equations. Although such approaches are rather heavy,
costly, and mask some interesting physical features, they are useful as they
may be regarded as careful numerical control of the semi-analytical methods
and associated models.
6.3 Thermal Convection
Fluid motion driven by thermal gradients, also called thermal convection, is
a familiar and important process in nature. It is far from being an academic
subject. Beyond its numerous technological applications, it is the basis for
the interpretation of several phenomena as the drift of the continental plates,
144 6 Instabilities and Pattern Formation
the Sun activity, the large-scale circulations observed in the oceans, the at-
mosphere, etc. As a prototype of thermal convection, we shall examine the
behaviour of a thin fluid layer enclosed between two horizontal surfaces whose
lateral dimensions are much larger than the width of the layer. The two hori-
zontal bounding planes are either rigid plates or stress-free surfaces, the lower
surface is uniformly heated so that the fluid is subject to a vertical tempera-
ture gradient. If the temperature gradient is sufficiently small, heat is trans-
ferred by conduction alone and no motion is observed. When the temperature
difference between the two plates exceeds some critical value, the conduction
state becomes unstable and motion sets in. The most influential experimental
investigation on thermal convection dates back to B´enard (1900). The fluid
used by B´enard was molten spermaceti, a whale’s non-volatile viscous oil, and
the motion was made visible by graphite or aluminium powder. In B´enard’s
original experiment, the lower surface was a rigid plate but the upper one
was open to air, which introduces an asymmetry in the boundary conditions
besides surface tension effects. The essential result of B´enard’s experiment
was the occurrence of a stable, regular pattern of hexagonal convection cells.
Further investigations showed that the flow was ascending in the centres of
the cells and descending along the vertical walls of the hexagons. Moreover,
optical investigations revealed that the fluid surface was slightly depressed at
the centre of the cells.
A first theoretical interpretation of thermal convection was provided by
Rayleigh (1920), whose analysis was inspired by the experimental observa-
tions of B´enard. Rayleigh assumed that the fluid was confined between two
free perfectly heat conducting surfaces, and that the fluid properties were
constant except for the mass density. In Rayleigh’s view, buoyancy is the
single responsible for the onset of instability. By assuming small infinitesimal
disturbances, he was able to derive the critical temperature gradient for the
onset of convection together with the wave number for the marginal mode.
However, it is presently recognized that Rayleigh’s theory is not adequate to
explain the convective mechanism investigated by B´enard. Indeed in B´enard’s
set up, the upper surface is in contact with air, and surface tractions originat-
ing from surface tension gradients may have a determinant influence on the
onset of the flow. By using stress-free boundary conditions, Rayleigh com-
pletely disregarded this effect. It should also be realized that surface tension
is not a constant but that it may depend on the temperature or (and) the
presence of surface contaminants. This dependence is called the capillary or
the Marangoni effect after the name of the nineteenth-century Italian investi-
gator. The importance of this effect was only established more than 40 years
later after Rayleigh’s paper by Block (1956) from the experimental point of
view. Pearson (1958) made the first theoretical study about the influence of
the variation of surface tension with temperature on thermal convection. The
predominance of the Marangoni effect in B´enard’s original experiment is now
admitted beyond doubt and confirmed by experiments conducted recently in
space-flight missions where gravity is negligible. When only buoyancy effects
6.3 Thermal Convection 145
are accounted for, the problem is generally referred to as Rayleigh–B´enard’s
instability while B´enard–Marangoni is the name used to designate surface
tension-driven instability. When both buoyancy and surface tension effects
are present, one speaks about the Rayleigh–B´enard–Marangoni’s instability.
6.3.1 The Rayleigh–B´enard’s Instability:
A Linear Theory
We are going to study the instabilities occurring in a viscous fluid layer of
thickness d (between a few millimetres and a few centimetres) and infinite
horizontal extent limited by two horizontal non-deformable free surfaces, the
z-axis is pointing in the opposite direction of the gravity acceleration g. The
fluid is heated from below with T
h
and T
c
, the temperatures of the lower and
upper surfaces, respectively (see Fig. 6.4). The mass density ρ is assumed to
decrease linearly with the temperature according to the law
ρ = ρ
0
[1 −α(T −T
0
)], (6.18)
where T
0
is an arbitrary reference temperature, say the temperature of the
laboratory, and α the coefficient of thermal expansion, generally a positive
quantity except for water around 4

C. For ordinary liquids, α is of the order
of 10
−3
–10
−4
K
−1
.
When the temperature difference ∆T = T
h
−T
c
(typically not more than
a few

C) between the two bounding surfaces is lower than some critical
value, no motion is observed and heat propagates only by conduction inside
the fluid. However by further increasing ∆T, the basic heat conductive state
becomes unstable at a critical value (∆T)
crit
and matter begins to perform
bulk motions which, in rectangular containers, take the form of regular rolls
aligned parallel to the short side as visualized in Fig. 6.5, this structure is
referred to as a roll pattern. Note that the direction of rotation of the cells is
unpredictable and uncontrollable, and that two adjacent rolls are rotating in
opposite directions.
Fig. 6.4 Horizontal fluid layer submitted to a temperature gradient opposed to the
acceleration of gravity g
146 6 Instabilities and Pattern Formation
Fig. 6.5 Convective rolls in Rayleigh–B´enard’s instability
A qualitative interpretation of the onset of motion is the following. By
submitting the fluid layer to a temperature difference, one generates a tem-
perature and a density gradient. A fluid droplet close to the hot lower plate
has a lower density than everywhere in the layer, as density is generally a
decreasing function of temperature. As long as it remains in place, the fluid
parcel is surrounded by particles of the same density, and all the forces acting
on it are balanced. Assume now that, due to a local fluctuation, the droplet is
slightly displaced upward. Being surrounded by cooler and denser fluid, it will
experience a net upward Archimede’s buoyant force proportional to its vol-
ume and the temperature difference whose effect is to amplify the ascending
motion. Similarly a small droplet initially close to the upper cold plate and
moving downward will enter a region of lower density and becomes heavier
than the surrounding particles. It will therefore continue to sink, amplifying
the initial descent. What is observed in the experiments is thus the result
of these upward and downward motions.
However, experience tells us that convection does not appear whatever
the temperature gradient as could be inferred from the above argument. The
reason is that stabilizing effects oppose the destabilizing role of the buoyancy
force; one of them is viscosity, which generates a friction force directed oppo-
site to the motion, the second one is heat diffusion, which tends to spread out
the heat contained in the droplet towards its environment reducing the tem-
perature difference between the droplet and its surroundings. This explains
why a critical temperature difference is necessary to generate a convective
flow: motion will start as soon as buoyancy overcomes the dissipative effects
of viscous friction and heat diffusion. These effects are best quantified by the
introduction of the thermal diffusion time and the viscous relaxation time
τ
χ
= d
2
/χ, τ
ν
= d
2
/ν, (6.19)
where χ is the thermal diffusivity, ν the kinematic viscosity, and d a scaling
length, τ
χ
is the time required by the fluid to reach thermal equilibrium
with its environment, τ
ν
is related to the time needed to obtain mechanical
6.3 Thermal Convection 147
equilibrium. Another relevant timescale is the buoyant time, i.e. the time that
a droplet, differing from its environment by a density defect δρ = ρ
0
α∆T,
needs to travel across a layer of thickness d,
τ
2
B
= d/(αg∆T). (6.20)
This result is readily derived from Newton’s law of motion ρ
0
d
2
z/dt
2
= gδρ
for a small volume element; a large value of ∆T means that the buoyant
time is short. To give an order of magnitude of these various timescales,
let us consider a shallow layer of silicone oil characterized by d = 10
−3
m,
ν = 10
−4
m
2
s
−1
, χ = 10
−7
m
2
s
−1
, it is then found that τ
ν
= 10
−2
s and
τ
χ
= 10 s.
The relative importance of the buoyant and dissipative forces is obtained
by considering the ratios τ
ν

B
and τ
χ

B
or, since they occur simultaneously,
through the so-called dimensionless Rayleigh number,
Ra =
τ
ν
τ
χ
τ
2
B
=
αg∆Td
3
νχ
. (6.21)
The Rayleigh number can therefore be viewed as the ratio between the desta-
bilizing buoyancy force and the stabilizing effects expressed by the viscous
drag and the thermal diffusion; convection will start when Rayleigh number
exceeds some critical value (Ra)
c
. For Ra < (Ra)
c
, the fluid remains at rest
and heat is only transferred by conduction, for Ra > (Ra)
c
, there is a sudden
transition to a complex behaviour characterized by the emergence of order in
the system. The ratio between the dissipative processes is measured by the
dimensionless Prandtl number defined as
Pr = τ
χ

ν
= ν/χ, (6.22)
for gases Pr ∼ 1, for water Pr = 7, for silicone oils Pr is of the order of 10
3
,
and for the Earth’s mantle Pr ∼ 10
23
.
In a linear stability approach, the main problem is the determination of
the marginal stability curve, i.e. the curve of Ra vs. the wave number k at
σ = 0. The one corresponding to Rayleigh–B´enard’s instability is derived in
the Box 6.1.
Box 6.1 Marginal Stability Curve
The mathematical analysis is based on the equations of fluid mechanics writ-
ten within the Boussinesq approximation. This means first that the density
is considered to be constant except in the buoyancy term; second that all the
material properties as viscosity, thermal diffusivity, and thermal expansion
coefficient are temperature independent; and third that mechanical dissi-
pated energy is negligible. The governing equations of mass, momentum,
and energy balance are then given by
∇ v = 0,
148 6 Instabilities and Pattern Formation
_
∂v
∂t
+v ∇v
_
= −
1
ρ
0
∇p + [1 −α(T −T
0
)]g +ν∇
2
v, (6.1.1)
∂T
∂t
+v ∇T = χ∇
2
T,
where use has been made of the equation of state (6.18) and where
v(v
x
, v
y
, v
z
) and p designate the velocity and pressure fields, respectively.
The set (6.1.1) represents five scalar partial differential equations for the
five unknowns p, T and v
x
, v
y
, v
z
.
In the basic unperturbed state, the fluid is at rest and temperature is
conveyed by conduction, so that the solutions of (6.1.1) are simply
v
r
= 0, T
r
= T
h

∆T
d
z,
∂p
r
∂z
= −ρ
0
g[1 −α(T
r
−T
0
)], (6.1.2)
where subscript r refers to the unperturbed reference solution, z denotes the
vertical coordinate measured positive upwards with z = 0 corresponding to
the lower boundary and x, y the horizontal coordinates, and ∆T = T
h
−T
c
is the positive temperature difference between the lower and upper bound-
aries. Designating by v

= v −0, T

= T −T
r
, p

= p−p
r
the infinitesimally
small perturbations of the basic state, we can linearize (6.1.1) and obtain
the following set for the perturbed fields
∇ v

= 0, (6.1.3)
∂v

∂t
= −
1
ρ
0
∇p

+αT

ge
z
+ν∇
2
v

, (6.1.4)
∂T

∂t
+v

∇T
r
= χ∇
2
T

, (6.1.5)
where e
z
is the unit vector pointing opposite to g. We now determine the
corresponding boundary conditions. If we assume that the thermal conduc-
tivity at the limiting surfaces is much higher than in the fluid itself, any
thermal disturbance advected by the fluid will be instantaneously smoothed
out so that T

will vanish at the bounding surfaces. Since the horizontal
boundaries are assumed to be free surfaces, the shearing stress is zero at
the surface; when use is made of the equation of continuity (6.1.3), this
condition is identical to setting ∂
2
v

z
/∂z
2
= 0 together with v

z
= 0, as the
surfaces are non-deformable. Summarizing, the boundary conditions are
v

z
= ∂
2
v

z
/∂z
2
= 0, T

= 0 at perfectly heat conducting free surfaces.
(6.1.6)
In the case of rigid boundaries, the no-slip condition imposes that all the
components of the velocity are zero v

x
= v

y
= v

z
= 0, which combined with
the continuity condition yields
v

z
= ∂v

z
/∂z = 0, T

= 0 at perfectly heat conducting rigid walls.
(6.1.7)
6.3 Thermal Convection 149
At adiabatically isolated walls, the condition T

= 0 will be replaced by
∂T

/∂z = 0, expressing that the surface is impermeable to heat flow. When
the boundary surface is neither perfectly heat conducting nor adiabatically
isolated, heat transfer is governed by Newton’s cooling law −λ∂T/∂z =
h(T −T

) where λ is the heat conductivity of the fluid, h the so-called heat
transfer coefficient, and T

the temperature of the outside world.
A further simplification of the set (6.1.3)–(6.1.5) is obtained by applying
twice the operator rot (≡ ∇) on the momentum equation and using the
continuity equation, we are then left with the two following relations in the
two unknowns v

z
and T


∂t
(∇
2
v

z
) = αg∇
2
1
T

+ν∇
4
v

z
, (6.1.8)
∂T

∂t
= βv

z
+χ∇
2
T

, (6.1.9)
where ∇
2
1
≡ ∂
2
/∂x
2
+ ∂
2
/∂y
2
denotes the horizontal Laplacian whereas
β = ∆T/d. We now make the variables dimensionless by introducing the
following scaling
X =
x
d
, Y =
y
d
, Z =
z
d
,
ˆ
t =
ν
d
2
t, w =
d
ν
v

z
, θ =
χ
ν
T

∆T
, (6.1.10)
and solve (6.1.8) and (6.1.9) with a normal mode solution of the form
w = W(Z) exp[i(k
x
X +k
y
Y )] exp(σ
ˆ
t), (6.1.11)
θ = Θ(Z) exp[i(k
x
X +k
y
Y )] exp(σ
ˆ
t), (6.1.12)
where W(Z) and Θ(Z) are the amplitudes of the perturbations, k
x
and k
y
are the dimensionless wave numbers in the directions x and y, respectively,
and σ is the dimensionless growth rate. Since the present problem is self-
adjoint, it can be proved (Chandrasekhar 1961) that σ is a real quantity.
There exist several ways to make the variables dimensionless, the choice
made here seems to be one of the most preferred. Sometimes, it is also
preferable to describe the horizontal periodicity of the flow by means of the
wavelength λ = 2π/k rather than the wave number as it provides a direct
measure of the dimensions of the cells.
Substitution of solutions (6.1.11) and (6.1.12) in (6.1.8) and (6.1.9)
leads to the following amplitude differential equations
(D
2
−k
2
)(D
2
−k
2
−σ)W = Ra k
2
Θ, (6.1.13)
(D
2
−k
2
−σ Pr)Θ = −W, (6.1.14)
where D stands for d/dZ and k
2
= k
2
x
+k
2
y
, Ra and Pr denote the dimen-
sionless Rayleigh and Prandtl numbers defined by (6.21) and (6.22). The
boundary conditions corresponding to two perfectly heat conducting free
surfaces are
150 6 Instabilities and Pattern Formation
W = D
2
W = 0, Θ = 0 at Z = 0 and Z = 1. (6.1.15)
The latter are satisfied for solutions of the form W = Asin πZ, Θ =
Bsin πZ, which substituted in (6.1.13) and (6.1.14) lead to the following
algebraic equations for the two unknowns A and B:
A+ (π
2
+k
2
+σ)B = 0, (6.1.16)

2
+k
2
)(π
2
+k
2
+σPr
−1
)A−k
2
Ra B = 0.
Non-trivial solutions demand that the determinant of the coefficients
vanishes, which results in the following dispersion relation between k, σ,
Ra, and Pr:

2
+k
2

2
+ (π
2
+k
2
)
2
(1 +Pr)σ +Pr[(π
2
+k
2
)
3
−k
2
Ra] = 0. (6.1.17)
By setting σ = 0 in (6.1.17), one obtains the marginal curve (Ra)
0
vs. k
determining the Rayleigh number at the onset of convection (Fig. 6.6); it is
independent of the Prandtl number and given by
(Ra)
0
=

2
+k
2
)
3
k
2
. (6.23)
Fig. 6.6 Marginal stability curve for Rayleigh–B´enard’s instability in a horizontal
fluid layer limited by two stress-free perfectly heat conducting surfaces
6.3 Thermal Convection 151
The solutions of (6.1.17) can be written as
σ =

2
+k
2
)(1 +Pr)
2
_
−1 ±
_
1 +
4(Ra −(Ra)
0
)
(Ra)
0
(1 +Pr)(1 +Pr
−1
)
_
1/2
_
,
(6.24)
which shows explicitly that the growth rate is a function of k, Ra, and Pr.
For Ra < (Ra)
0
, one has σ < 0 (stability) while for Ra > (Ra)
0
, σ > 0
(instability). It follows also from (6.24) that σ is real when Ra > 0 and
that σ < 0 when Ra < 0; this latter result indicates that the system is
unconditionally stable by heating from above.
For slightly supercritical conditions, where Ra is close to (Ra)
0
, (6.24)
reads as
σ = (π
2
+k
2
)
Ra −(Ra)
0
(Ra)
0
(1 +Pr
−1
)
, (6.25)
which shows that, for Ra larger than (Ra)
0
, the amplitude of the disturbances
amplifies exponentially and the basic state is unstable. However, for such
values, the non-linear terms become important and the linear analysis ceases
to be valid.
The minimum value of the marginal curve is obtained from relation (6.23)
by differentiation with respect to k; setting this result equal to zero gives the
critical wave number k
c
at which the curve (Ra)
0
(k) is minimum, and the
corresponding critical Rayleigh number (Ra)
c
. For the present problem, it is
found that
k
c
= π/

2 = 2.21, (Ra)
c
= 27π
4
/4 = 657.5 (free–free boundary conditions).
(6.26)
The critical Rayleigh number allows us to determine the critical temperature
difference at which the system changes from the state of rest to the state
of cellular motion, the critical wave number provides information about the
horizontal periodicity of the patterns at the onset of convection, k
c
represents
the most dangerous mode picked up by the fluid. For other boundary condi-
tions, the calculations are more complicated but the procedure remains valid,
the marginal curves will have approximately the same form as in Fig. 6.6 with
critical values given by
k
c
= 2.68, (Ra)
c
= 1, 100.6 (rigid–free boundaries),
k
c
= 3.117, (Ra)
c
= 1, 707.7 (rigid–rigid boundaries).
As expected, stability is reinforced (larger (Ra)
c
value) in presence of rigid
surfaces as the fluid motion is more strongly inhibited by the viscous forces.
At the same time the dimensions of the cells (larger k
c
value) are diminished:
more energy must now be dissipated to compensate for the larger release of
energy by buoyancy, clearly narrower cells are associated with greater dis-
sipation and energy release. It is also worth to note that, for Ra > (Ra)
c
152 6 Instabilities and Pattern Formation
and according to the linear theory, a continuous spectrum of modes becomes
unstable from which follows that the observed pattern should be very intri-
cate. This is in contradiction with the observation that the flow prefers rather
simple cellular forms; the reason for this discrepancy must once more be at-
tributed to the omission of non-linear terms. Moreover, the linear theory is
unable to predict the particular pattern (either rolls, squares, or hexagons)
selected by the fluid; this is so because the eigenvalue problem is degenerate,
which means that to one eigenvalue Ra there corresponds an infinite number
of possible patterns with the same wave number k. The reason why a partic-
ular pattern is selected can only be understood from a non-linear approach,
which shows that, for the present problem, two-dimensional parallel rolls are
the preferred patterns as confirmed by experimental observations.
6.3.2 The Rayleigh–B´enard’s Instability:
A Non-Linear Theory
We now examine the behaviour of the amplitude of the disturbance beyond
the critical point. Returning to the linear theory where the amplitude is
supposed to behave as A(t) ∼ exp σt, we may write that, at threshold, the
relevant differential equation is
dA
dt
= σA, (6.27)
where σ is given by (6.25). For supercritical Rayleigh numbers, the amplitude
will then increase exponentially but non-linear self-interaction between modes
becomes important giving raise to higher-order terms in A
n
. We are then led
to an amplitude equation of the form suggested by Landau, i.e.
dA
dt
= σA−lA
3
, (6.28)
where l is a positive constant to be determined from the boundary conditions
whereas σ is the growth rate corresponding to the most dangerous mode k
c
(= π/

2 for free–free boundaries). In virtue of (6.25) and for large values of
Pr as in silicone oils, σ is given by
σ = 3k
2
c
ε, (6.29)
in which
ε = [Ra −(Ra)
c
]/(Ra)
c
(6.30)
measures the relative distance from the critical point. There is no term in A
2
in (6.28) because it is assumed that the convective pattern is such that, by
reversing the fluid velocities, the same pattern is observed; this implies that
(6.28) must be invariant with respect to the symmetry A = −A. Of course
6.3 Thermal Convection 153
this is only true for two-dimensional rolls and it is not so for three-dimensional
patterns as hexagonal cells. Solving (6.28) for steady conditions, one obtains
A
s
= (σ/l)
1/2
= k
c
(3ε/l)
1/2
, (6.31)
from which it follows that the steady amplitude A
s
is proportional to ε
1/2
; this
result is typical of a supercritical bifurcation as represented on Fig. 6.1. To
examine the stability of the steady solution, let us superpose an infinitesimal
disturbance A

such that A = A
s
+A

. Substituting this expression in (6.28)
and using dA
s
/dt = 0 leads to
dA

dt
= −2σA

= −6k
2
c
εA

, (6.32)
and, after integration,
A

∼ exp[−6k
2
c
εt]. (6.33)
The perturbation decays for ε > 0, i.e. for Ra > (Ra)
c
, and it is concluded
that two-dimensional rolls are stable for perturbations with a wave number
equal to the critical one. When several modes are taken into account as in
the B´enard–Marangoni’s problem which is treated in Sect. 6.3.3, the problem
is much more complicated because various modes will interact and instead of
(6.28), we are led to a coupled set of non-linear equations of the form
dA
n
dt
= σA
n
−A
n

m
l
mn
A
2
m
−lA
3
n
(m, n = 1, 2, . . .). (6.34)
An example of such kind of equations is the famous Lorenz model (1963),
which is widely used in meteorology and which is at the essence of the theory
of chaos. The explicit form of Lorenz set of equations is
dA
1
dt
= Pr(A
2
−A
1
),
dA
2
dt
= rA
1
−A
2
−A
1
A
3
,
dA
3
dt
= −bA
3
+A
1
A
2
,
(6.35)
where r = Ra/(Ra)
c
and b = 4/(1 + 4k
2
). The coefficient A
1
describes the
velocity of the fluid particles, A
2
the temperature fluctuations, and A
3
a
horizontally averaged temperature mode. A detailed discussion of the Lorenz
model can, for example, be found in Sparrow (1982).
When the presence of lateral walls is taken into account, the amplitude
equations must include extra terms expressing the variation of the amplitudes
with respect to the horizontal coordinates. An abbreviated version is the
following
∂A
∂t
= σA−lA
3


2
A
∂x
2
, (6.36)
which is called the Ginzburg–Landau’s equation in reference to a paper by
these authors on superconductivity, wherein an equation of the above form
was given.
154 6 Instabilities and Pattern Formation
The main important result of the above analysis is that the infinite hor-
izontal plane should be covered with straight parallel rolls of infinite length
and oriented randomly. The randomness of the pattern finds its origin in
the randomness of the initial disturbances in an infinite plane theory. This
degeneracy may be overcome either by enclosing the fluid in boxes of finite
aspect ratio or by using a grid with an imposed spacing. When the grid is
slightly heated by a lamp and placed above the upper cooled surface, it in-
duces rolls whose wavelength corresponds to the grid spacing. By increasing
the Rayleigh number, one can maintain these rolls and follow their evolution
in the course of time.
6.3.3 B´enard–Marangoni’s Surface Tension-Driven
Instability
In Sect. 6.3.2, we have shown that, in sufficiently deep fluid layers, convection
sets in when buoyancy overcomes heat dissipation and viscous forces, and that
only rolls are stable for small and moderate Rayleigh numbers. Alternatively,
in rather shallow layers with a upper surface open to air, as in B´enard’s
original experience, a regular pattern of hexagons is observed as shown in
Fig. 6.7.
It is now admitted that it is the temperature dependence of the surface
tension at interface, which is responsible for the hexagonal flow pattern.
Pearson (1958) was the first to propose a theoretical analysis on the influence
of the surface tension o assumed to depend linearly on the temperature
Fig. 6.7 B´enard’s hexagonal pattern viewed from above (from Koschmieder 1993)
6.3 Thermal Convection 155
o = o
0
+
∂o
∂T
(T −T
0
), (6.37)
wherein ∂o/∂T is usually a negative quantity. The linear analysis is similar
to that of the Rayleigh–B´enard’s problem, but the dimensionless Rayleigh
number is now replaced by the Marangoni number
Ma = −
∂o
∂T
∆T
d
ρ
0
νχ
, (6.38)
where ∆T is the temperature difference applied between the boundaries of
the fluid layer. At the upper surface, the boundary conditions read now, in
dimensionless form,
W = 0, D
2
W = k
2
Ma Θ (D = d/dZ). (6.39)
The first relation arises from the assumed non-deformability of the surface,
whereas the second one expresses the equality between the normal compo-
nent of the mechanical stress tensor and the horizontal gradient of the surface
tension (see Problem 6.8). For a stress-free boundary, the Marangoni num-
ber vanishes and one finds back the classical result D
2
W = 0. It is impor-
tant to realize that there are significant differences between the Rayleigh
and Marangoni numbers, as the former varies as d
3
and the latter as d;
this means that surface tension-driven instability will be dominant in thin
layers (small d), whereas buoyancy effects will be more important in thick
layers (large d). Remembering that in B´enard’s original experiments the
depth of the layer was about 1 mm, it is not surprising that the instabil-
ity observed by B´enard was essentially caused by surface tension forces. If
the ratio Ma/Ra →∞, the surface forces dominate the dynamics of the fluid
and gravity effects can be neglected. Under terrestrial conditions, Ma/Ra = 4
for a layer with a thickness d = 1 mm and Ma/Ra = 4 10
−4
for d = 10 cm.
Under microgravity conditions (g ≈ 0), Ma/Ra = 10
4
for d = 1 mm.
As an illustration, consider a fluid layer (in which buoyancy effects are
neglected, i.e. Ra = 0) in contact with a perfectly heat conducting bot-
tom plate and whose upper surface is adiabatically isolated but subject to a
temperature-dependent surface tension. The marginal stability curve giving
(Ma)
0
as a function of k is similar to the marginal curve of the Rayleigh–
B´enard’s problem (Pearson 1958)
(Ma)
0
=
8k
2
cosh k(k −sinh k cosh k)
k
3
cosh k −sinh
3
k
, (6.40)
and is represented in Fig. 6.8. The corresponding critical values are (Ma)
c
=
79.61 and k
c
= 1.99.
The theory has been advanced further by Scriven and Sternling (1964) to
allow for deformations of the free surface, in which case the onset of convec-
tion may be oscillatory. Nield (1964) extended Pearson’s analysis to account
for coupled buoyancy and surface tension effects. The numerical results (see
156 6 Instabilities and Pattern Formation
Fig. 6.8 Marginal stability curves for coupled buoyancy and surface tension-driven
instability; the curve Ra = 0 corresponds to surface tension-driven instability
Fig. 6.8) show that the critical value (Ma)
c
decreases upon increasing Ra
but that the critical wave number remains practically unchanged. Nield was
able to derive the following linear relation between the critical Rayleigh and
Marangoni numbers
Ra
(Ra)
c
+
Ma
(Ma)
c
= 1, (6.41)
which exhibits the strong coupling between the two motors of instability;
(Ra)
c
and (Ma)
c
are the critical values corresponding to the absence of sur-
face tension gradients and gravity, respectively.
All the above theoretical considerations are based on linearized equations;
they give interesting information on the onset of convection but are silent
about the planform and other characteristics of the flow; to go further a
non-linear analysis is required.
Non-linear effects were considered by Cloot and Lebon (1984), who solved
the problem with a power series expansion (Shl¨ uter et al. 1965), and by
Bragard and Lebon (1993) using the amplitude method. To incorporate pos-
sible hexagonal patterns, the velocity disturbance is taken to be of the general
form
w = W(z)
_
Z(t) cos ky +Y (t) cos(

3/2kx) cos(ky/2)
_
, (6.42)
6.3 Thermal Convection 157
which corresponds to rolls for Y = 0 and to hexagons for Y = 2Z. After some
lengthy but rather simple calculations, Bragard and Lebon (1993) obtained
under zero-gravity conditions the following set of amplitude equations
dY
dt
= εY −γY Z −RY
3
−PY Z
2
,
dZ
dt
= εZ −
1
4
γY
2
−R
1
Z
3

1
2
PY
2
Z,
(6.43)
where ε = [Ma−(Ma)
c
]/(Ma)
c
and γ, P = 4R−R
1
, R, and R
1
are constant
coefficients determined from the solvability condition. Observe that the sym-
metry property f(Y ) = f(−Y ) and f(Z) = f(−Z) is broken in (6.43). To
obtain a hexagonal structure, the condition Y = 2Z must be fulfilled and it
is directly checked that the steady versions of (6.43) possess a solution ver-
ifying this condition. By performing a linear stability analysis of the steady
solutions, it was shown (Bragard and Lebon 1993) that hexagons are stable
in the range −0.0056 < ε < 1.8. By setting Y = 2Z in the stationary solution
of (6.43), one obtains a second-order equation in Z, namely
ε −γZ −(R
1
+ 2P)Z
2
= 0, (6.44)
which is represented in Fig. 6.9 and displays an inverted bifurcation, confirm-
ing the presence of a region of subcritical instability, characterized by the
occurrence of motion at ε < 0.
Experiments by Koschmieder (1993) indicate that the pattern of the sub-
critical instability consists of ill-defined small-scale hexagonal cells. For the
sake of completeness, we have also reported in Fig. 6.9 the amplitude curve
for the roll solution. The solid portions of the curves refer to stable mo-
tions and the broken lines to unstable solutions. For ε < ε

(= −0.0056),
Fig. 6.9 Amplitude Z vs. ε: solid lines correspond to stable solutions and broken
lines correspond to unstable solutions
158 6 Instabilities and Pattern Formation
the fluid remains at rest, as soon as ε is larger than ε

but less than zero,
subcritical stable hexagons are predicted. Moreover, hexagons are the only
stable solutions in the interval ε

< ε < ε
1
(= 0.53) while in the range
ε
1
< ε < ε
2
(= 1.8), both hexagonal cells and rolls are stable; finally for
ε > ε
2
, only stable rolls can be found.
A behaviour similar to B´enard–Marangoni’s instability is observed in
Rayleigh–B´enard’s problem when Boussinesq approximation is given up. In
the particular case of a temperature-dependent viscosity, it was found (Palm
1975) that the hexagonal pattern is the preferred structure, even for a fluid
enclosed between two rigid plates. Experiments have also revealed that the
motion in the hexagonal platform may be either ascending or descending at
the centre of the cell: the motion is downward in the cell centres of gases and
upward in liquids. This change in circulation is associated to the property
that, in gases, the viscosity increases with temperature while it decreases in
liquids.
Other interesting aspects of buoyancy and surface tension-driven instabil-
ities have been treated in the specialized literature (e.g. Colinet et al. 2002).
Among them, let us mention the problem of fluid layers submitted to hori-
zontal temperature gradients. This is important in crystal growth processes
(Davis 1987; Parmentier et al. 1993; Madruga et al. 2003), convection in
porous media, which is of interest in the mineral oil industry, thermal con-
vection in rotating systems (Busse 1978; Davis 1987), and double diffusive
convection under the mutual action of temperature and salinity gradients
(thermohaline convection and salt fingers). All these phenomena play a cen-
tral role in geophysics, astrophysics, and oceanography. Let us also point out
the effects resulting from surface deformations (Regnier et al. 2000), partic-
ularly relevant in small boxes and in presence of lateral walls, unavoidable in
practical experiments (Rosenblat et al. 1982; Dauby and Lebon 1996; Dauby
et al. 1997). B´enard’s convection has also been used to model the Earth’s
mantle motions (Turcotte 1992); moreover, it offers an attractive basis for the
interpretation of the transition to turbulence (Berg´e et al. 1984). The above
list, although being not exhaustive, reflects the impressively wide range of
applications of Rayleigh–B´enard–Marangoni’s instability.
6.4 Taylor’s Instability
Taylor’s instability is observed in a viscous incompressible fluid column con-
tained between two vertical cylinders rotating at different angular velocities,
the temperature being assumed to remain uniform. There occurs a compe-
tition between centrifugal forces and viscosity: if we assume that the inner
cylinder moves at a higher velocity than the outer one, the fluid close to
the inside wall will move outward and replace there the slow moving fluid.
At small angular velocity, one has an ordinary Couette flow where angular
6.4 Taylor’s Instability 159
Fig. 6.10 (a) Taylor’s vortices, (b) wavy flow, and (c) turbulence (from Coles 1965)
momentum is transported from the inner to the outer cylinder by viscosity
in the form of stationary annular convective cells. When the angular veloc-
ity is gradually increased, this state becomes unstable and toroidal vortices,
also called Taylor’s vortices, are formed. At higher angular velocities, wavy
deformations appear and finally a transition to turbulent motion occurs (see
Fig. 6.10).
At low angular velocity, the motion is laminar with a velocity field given
by
v
r
= 0, v
θ
= Ar +
B
r
, v
z
= 0, (6.45)
when expressed in cylindrical coordinates r, θ, z, where r is the radial distance
to the axis of the cylinders, θ the azimuthal angle, and z the coordinate
parallel to the vertical axis, A and B are constants to be determined from
the boundary conditions. With Ω
1
and Ω
2
the angular velocities of the inner
and outer cylinders, respectively, and R
1
and R
2
the corresponding radii, the
non-slip condition imposes the following conditions at the walls
v
θ
= R
1

1
at r = R
1
, v
θ
= R
2

2
at r = R
2
; (6.46)
this allows us to determine the values of the constants A and B, namely
A = −Ω
1

2
−µ
1 −Ω
2
, B = Ω
1
R
2
1
(1 −µ)
1 −Ω
2
, (6.47)
where Ω = Ω
2
/Ω
1
and µ = R
1
/R
2
.
We first proceed with a linear stability analysis. Denote the velocity dis-
turbance by u

, v

, w

and the pressure disturbance by p

; assuming that
160 6 Instabilities and Pattern Formation
the disturbances are small and axisymmetric, i.e. θ independent, it is easily
checked that they obey the following linearized continuity and Navier–Stokes
equations
1
r

∂r
(ru

) +
∂w

∂z
= 0, (6.48a)
∂u

∂t
−2v
θ
v

r
= −

∂r
_
p

ρ
_
−ν
_

2
u


u

r
2
_
, (6.48b)
∂v

∂t
+
_
∂v
θ
∂r
+
v
θ
r
_
u

= ν
_

2
v


v

r
2
_
, (6.48c)
∂w

∂t
= −

∂z
_
p

ρ
_
+ν∇
2
w

, (6.48d)
where ∇
2
≡ (1/r)(∂/∂r) + ∂
2
/∂r
2
+ ∂
2
/∂z
2
. The pressure disturbance p

is
easily eliminated from relations (6.48b) and (6.48d). Moreover, by suppos-
ing that the fluid column is of infinite length, we need only the boundary
conditions at R
1
and R
2
, namely
u

= v

= w

= 0. (6.49)
According to the normal mode technique, we seek for solutions of the form
(u

, v

, w

) = [U(r), V (r), W(r)] exp(ikz +σt). (6.50)
To simplify the calculations, we shall make the narrow-gap approximation,
which implies that the distance d = R
2
− R
1
between the cylinders is much
smaller than R
1
. This permits us to put µ = 1 and to omit 1/r compared to
∂/∂r, moreover within the small-gap limit, A + B/r
2
≈ Ω
1
and A ≈ −Ω
1
.
Substituting (6.50) in (6.48), making all the quantities dimensionless, and
proceeding by analogy with the Rayleigh–B´enard’s problem, we obtain the
following relations for the radial and azimuthal velocity amplitudes at the
marginal state σ = 0
(D
2
−k
2
)
2
U = [1 + (Ω −1)Z]V, (6.51)
(D
2
−k
2
)V = −Ta k
2
U, (6.52)
where we have introduced the notation D = d/dZ with
Z = (r −R
1
)/d (6.53)
and
Ta = 4Ω
2
1
d
4

2
, (6.54)
the relevant boundary conditions are
U = DU = V = 0 at Z = 0 and Z = 1. (6.55)
6.4 Taylor’s Instability 161
The quantity Ta is the dimensionless Taylor number, which is the ratio be-
tween the centrifugal forces and the viscous dissipation. It is worth to stress
that relations (6.51) and (6.52) are similar to (6.1.13) and (6.1.14) describing
Rayleigh–B´enard’s instability. For each particular value of Ω = Ω
2
/Ω
1
, the
eigenvalue problem (6.51) and (6.52) results in a marginal stability curve (see
Fig. 6.11), which is similar to the curves for the Rayleigh–B´enard’s problem
in the rigid–rigid configuration. The minimum of the curve gives the criti-
cal value (Ta)
c
determining the minimal angular velocity at which toroidal
axisymmetric vortices will set in. For Ta < (Ta)
c
, all the disturbances are
dampened and the flow is independent of the z coordinate; at Ta = (Ta)
c
,
there is a supercritical transition from Couette flow to Taylor’s vortices char-
acterized by a critical wave number k
c
or a critical wavelength λ
c
= 2π/k
c
which is easily detectable on the photographs of Fig. 6.10. In the small-gap
limit, the critical values are given by
(Ta)
c
= 3, 430/(1 +Ω), k
c
= 3.12, for 0 < Ω < 1. (6.56)
The critical Taylor number is a function of Ω and it is interesting to note
that, for Ω = 1, one has (Ta)
c
= 1, 715, which is very close to the value 1,708
obtained for the critical Rayleigh number in the rigid–rigid convection case.
The same remark applies to the critical wave number whose value k
c
= 3.12
is comparable to the value k
c
= 3.117 of the Rayleigh–B´enard’s problem. It is
therefore clear that the linear stability problems for Rayleigh–B´enard’s and
Taylor’s instabilities in the narrow-gap approximation are formally identical.
Fig. 6.11 Marginal stability curve expressing the Taylor number Ta as a function
of the wave number k
162 6 Instabilities and Pattern Formation
For wide distances between the cylinders, it is useful to work with a Taylor
number (Chandrasekhar 1961), which is an explicit function of the radius
ratio. The results remain qualitatively unchanged with respect to the small-
gap approximation; the main features are that the critical Taylor number
increases with decreasing values of the radius ratio whereas the critical wave
number remains practically unchanged.
To obtain the form of the fluid flow after the onset of instability, a non-
linear analysis is required. It is not our purpose to enter into the details of
the non-linear developments as they are rather intricate and therefore we
advice the interested reader to refer to the specialized literature for details.
A wide discussion can be found in Drazin and Reid (1981) and Koschmieder
(1993). Starting from an amplitude equation of the form (6.28), it is found
that, in the weakly supercritical non-linear regime, the amplitude of the flow
is proportional to [Ta −(Ta)
c
]
1/2
and the motion takes the form of toroidal
vortices, the interval of stable wave numbers lies, however, in a narrower band
than the one predicted by the linear approach. The stability of the Taylor’s
vortices was studied experimentally by Coles (1965). By increasing the Taylor
number, a sequence of bifurcations is displayed, which may be summarized
as follows. For Ta < (Ta)
c
, one has a pure Couette flow that becomes un-
stable at the critical Taylor number at which Taylor’s vortices are observed.
At higher values of Ta, wavy vortices with a characteristic frequency are dis-
played (see Fig. 6.10b), the next step is the appearance of a quasi-periodic
regime with two characteristic frequencies. Later on, a transition towards
weak turbulence characterized by the disappearance of one of the frequencies
takes place, still further, the second characteristic frequency vanishes and a
state of full turbulence emerges (see Fig. 6.10c). One of the interests of the
study of Taylor’s instability is that one can follow the transition from lam-
inar flow to turbulence in great detail through a limited number of rather
characteristic stages, it offers therefore an attractive route to an exhaustive
study of turbulence at the same footing as Rayleigh–B´enard’s convection.
In the forthcoming sections, we will discuss qualitatively several examples
of patterns occurring in such various domains as chemistry, biology, oceanog-
raphy, and electricity. Unlike the presentations of the B´enard–Marangoni’s
and Taylor’s instabilities, we shall not enter into detailed mathematical for-
mulations but rather outline the main features, referring the interested reader
to specialized books and articles.
6.5 Chemical Instabilities
Dissipative patterns are not exclusive of hydrodynamics but are also found in
many other processes and particularly in chemistry. Some chemical reactions
give rise to temporal and spatial variations of the mass concentrations of
the active substances. In the course of time, some substances may undergo
periodic oscillations whereas some spatial patterns may be formed in the
6.5 Chemical Instabilities 163
reaction vessel. It should be stressed that these phenomena are underlying
the activities of life and are therefore far from being purely academic games.
The central feature governing the appearance of self-organization in chemistry
is the autocatalytic reaction: it is a reaction in which the products of some
step take part to a subsequent step. It looks like a feedback process wherein
the presence of one particular species stimulates the production of more of
that species.
A representative example of chemical reaction exhibiting dissipative pat-
terns is the Belousov–Zhabotinsky (BZ) reaction, which is discussed later on.
When chemical reactions are coupled with diffusion, as it occurs in spatially
inhomogeneous mixtures, the corresponding spatio-temporal instabilities are
designated as Turing’s instabilities. In a remarkable paper, Turing (1952)
showed that a steady and uniform state in a reaction–diffusion system might
become unstable after some control parameter has overcome some critical value.
Maintenance of non-equilibrium states for a long period demands that
chemical reagents are added to the system. But a result of such a process
is that the concentrations of the various components are made very inho-
mogeneous and it is necessary to stir the mixture with the consequence
that spatial patterns will be destroyed. Only time oscillations can be ob-
served in such stirred homogeneous reactions. As illustrations, we will dis-
cuss in the next sections the Lotka–Volterra and the Brusselator models
(Prigogine and Lefever 1968; Nicolis and Prigogine 1977). To maintain non-
equilibrium conditions in absence of stirring has for a long time been
considered as a difficult problem. It is only recently that experimental meth-
ods using molecular diffusion have been developed which allow for the pres-
ence of both spatial and temporal patterns.
6.5.1 Temporal Organization in Spatially
Homogeneous Systems
6.5.1.1 The Lotka–Volterra Model
A simple example of temporal organization is the Lotka–Volterra model,
which is well known in ecology. It describes the growth and death of a pop-
ulation of X individuals of a species of prey and Y individuals of a species
of predators, X and Y represent for instance two populations of fishes where
Y is subsisting exclusively on X. This prey–predator system is governed by
the non-linear equations
dX
dt
= X(k
1
−k
2
Y ), (6.57)
dY
dt
= −Y (k
2
−k
3
X), (6.58)
where k
1
, k
2
, and k
3
are positive constants.
164 6 Instabilities and Pattern Formation
Relations (6.57) and (6.58) admit two trivial stationary solutions X
s
=
Y
s
= 0 and two non-trivial solutions
X
s
= k
2
/k
3
, Y
s
= k
1
/k
2
. (6.59)
Their linear stability is investigated by writing X = X
s
+x

, Y = Y
s
+y

and
by substituting these expressions in (6.57) and (6.58); making use of (6.59), it
is checked that the disturbances x

and y

obey the following linear equations
dx

dt
= −k
2
X
s
y

,
dy

dt
= k
3
Y
s
x

, (6.60)
and, after elimination of y

,
d
2
x

dt
2
= −k
1
k
2
x

, (6.61)
whose solution is of the form
x

= x
0
cos(ωt) with ω =
_
k
1
k
2
. (6.62)
It follows that the stationary state is not asymptotically stable in the sense
that the perturbations do not vanish in the limit of t →∞, instead they are
periodic in time with an angular frequency ω depending on the parameters
characterizing the system. This periodic behaviour is easy to interpret: when
the population of preys X is increased, the predators Y have more food
at their disposal and their number is increasing. Since Y consumes more
individuals X, their population diminishes and less food is available for the
species Y , which in turn decreases. This reduction of the number of predators
allows a larger number of preys to survive with the consequence that their
population will increase and a new cycle is initiated. The ratio of (6.58) and
(6.57) can be written as
dY
dX
= −
Y (k
2
−k
3
X)
X(k
1
−k
2
Y )
, (6.63)
which, after integration, yields the following relation
k
3
X +k
2
Y −k
2
log X −k
1
log Y = constant, (6.64)
with the constant depending on the initial conditions, and playing the same
role as total energy in classical mechanics. Clearly, (6.64) defines an infinity
of trajectories in the phase space X − Y corresponding to different initial
conditions (see Fig. 6.12).
These trajectories take the form of concentric curves surrounding the sta-
tionary state S with a period depending on the initial conditions. This be-
haviour is typical of conservative systems and has to be contrasted with limit
cycle oscillations as exhibited by dissipative systems. This gives also a strong
argument against the use of such models to describe oscillations observed in
nature, and particularly in some classes of chemical reactions.
6.5 Chemical Instabilities 165
Fig. 6.12 Phase space trajectories: the Lotka–Volterra cycles
6.5.1.2 The Brusselator
A more appropriate modelling of the oscillatory behaviour of chemical re-
actions is provided by the Brusselator (Prigogine and Lefever 1968; Nicolis
and Prigogine 1977) which consists in a simple and instructive example of
autocatalytic scheme. The sequence of chemical reactions in the Brusselator
is the following
A
k
1
−→X, (6.65a)
2X +Y
k
2
−→3X, (6.65b)
B +X
k
3
−→Y +D, (6.65c)
X
k
4
−→E, (6.65d)
the four steps are assumed to be irreversible which is achieved by taking
all reverse reaction constants equal to zero; the global reaction of the above
scheme is
A+B →D +E. (6.66)
The concentrations of the reactants A and B are maintained at a fixed and
uniform non-equilibrium value, and the final products D and E are removed
as soon as they are formed; the autocatalytic step is the second one, which
involves the intermediate species X and Y . It is assumed that the reac-
tions take place under isothermal and well stirred, i.e. spatially homogeneous
166 6 Instabilities and Pattern Formation
conditions. According to the laws of chemical kinetics, we have the following
rate equations for the species X and Y
dX
dt
= k
1
A+k
2
X
2
Y −k
3
BX −k
4
X, (6.67)
dY
dt
= −k
2
X
2
Y +k
3
BX, (6.68)
whose steady solutions are
X
s
=
k
1
k
4
A, Y
s
=
k
3
k
4
k
1
k
2
B
A
. (6.69)
To examine the stability of this solution, let us set X = X
s
+ x

and Y =
Y
s
+ y

where x

and y

are small disturbances; it is easy to verify that the
perturbations are obeying a linear system of the form
d
dt
_
x

y

_
= L
_
x

y

_
, (6.70)
where the matrix L is given by
L =


k
3
B −k
4
k
2
1
k
2
k
2
4
A
2
−k
3
B −
k
2
1
k
2
k
2
4
A
2


, (6.71)
whose eigenvalues σ
1
, σ
2
, . . . determine the stability of the system. It is left
as an exercise to show that the stationary state of (6.70) becomes unstable
(Re σ > 0) when B is larger than a critical concentration given by
B
c
=
k
4
k
3
+
k
2
1
k
2
k
3
k
2
4
A
2
. (6.72)
For B < B
c
, the system remains homogeneous; whereas at B = B
c
, the
eigenvalues are purely imaginary, which leads to undamped oscillations of
the perturbations, just like in the Lotka–Volterra model; finally for B > B
c
,
the system is unstable and the concentrations X and Y undergo periodic
oscillations of the limit cycle type, independently of the initial values of X and
Y . Such oscillations have indeed been detected in the Belousov–Zhabotinsky
reaction, where fascinating geometric patterns as concentric circles and spirals
propagating through the medium have been observed. BZ reaction is basically
a catalytic oxidation by potassium bromate KBrO of a organic compound
such as malonic acid CH
2
(COOH)
2
catalysed by the cerium Ce
3+
–Ce
4+
ion
couple.
After some transitory period, the oscillatory behaviour is evidenced by the
variation of the concentration of the Ce
3+
and Ce
4+
ions (Fig. 6.13). First,
a blue colour indicating an excess of Ce
4+
is spreading into the mixture, a
few minutes later the blue colour disappears and is replaced by a red one
6.5 Chemical Instabilities 167
Fig. 6.13 Experimentally observed oscillations of Ce
4+
and Ce
3+
ions in Belousov–
Zhabotinsky reaction
that indicates an excess of Ce
3+
ions. The process goes on with a succession
of blue, red, blue colours with a perfectly regular period, which to a certain
sense constitutes a chemical clock.
When the substances necessary for the reaction are exhausted, the oscil-
lations will die out and the system goes back to its equilibrium state. In
contrast with hydrodynamics where complexity is generally characterized by
non-homogeneities in space, in chemistry even a spatially homogeneous sys-
tem may exhibit a complex behaviour in time. More recent and detailed
analyses of BZ reaction have shown the occurrence of aperiodic and even
chaotic behaviours. There are several analogies of BZ periodic behaviour in
living organisms: heart beats, circadian rhythms, and menstrual cycles are a
few examples.
6.5.2 Spatial Organization in Spatially Heterogeneous
Systems
In absence of stirring, BZ reaction exhibits some non-trivial spatial patterns,
which arise from the interplay of the chemical reaction and the diffusion
process. When BZ reaction is performed in a long thin vertical tube so that
the problem consists of a single spatial dimension, one observes (Fig. 6.14)
a superposition of steady horizontal bands of different colours corresponding
to low and high concentration region. This structure is analogous to the
Rayleigh–B´enard’s and Taylor’s patterns.
6.5.3 Spatio-Temporal Patterns in Heterogeneous
Systems: Turing Structures
The great variety of patterns present in nature, both in the animate and
inanimate world, like the captivating beauty of a butterfly, the blobs of a
168 6 Instabilities and Pattern Formation
Fig. 6.14 Spatial structure in Belousov–Zhabotinsky reaction
leopard tail, or biological morphogenesis has been a subject of surprise and
interrogation for several generations of scientists. The question soon arises
about the mechanism behind them. In 1952, Turing proposed an answer based
on the coupling between (chemical) reactions and diffusion. As an example of
a Turing structure, let us still consider the Brusselator but we suppose now
that the chemical reaction takes place in a unstirred thin layer or a usual
vessel, so that spatial non-homogeneities are allowed. The basic relations are
the kinetic equations (6.67) and (6.68) to which are added the diffusion terms
D
x

2
X and D
y

2
Y , respectively, it is assumed that the diffusion coefficients
D
x
and D
y
are constant. By repeating the analysis of Sect. 6.5.1, it is shown
that the stationary homogeneous state becomes unstable for a concentration
B larger than the critical value
B
c
=
k
4
k
3
+
k
2
1
k
2
k
3
k
2
4
A
2
+
k
4
k
3
j
2
π
2
l
2
(D
x
+D
y
), j = 0, 1, 2 . . . (6.73)
at the condition that the diffusion coefficients are unequal, if D
x
= D
y
diffusion will not generate an instability, l is a characteristic length. Non-
homogeneities will begin to grow and stationary spatial patterns will emerge
in two-dimensional configurations. A rather successful reaction for observing
Turing’s patterns is the CIMA (chlorite–iodide–malonic acid) redox reaction,
which was proposed as an alternative to BZ reaction. The oscillatory and
space-forming behaviours in CIMA are made apparent through the presence
of coloured spots with a hexagonal symmetry. By changing the concentra-
tions, new patterns consisting of parallel narrow stripes are formed instead
of the spots (see Fig. 6.15).
Although it is intuitively believed that diffusion tends to homogenize
the concentrations, we have seen that, when coupled with an autocatalytic
6.6 Miscellaneous Examples of Pattern Formation 169
Fig. 6.15 Examples of Turing two-dimensional structures (from Vidal et al. 1994)
reaction under far from equilibrium conditions, it actually gives rise to spatial
structures. Turing’s stationary patterns are obtained when the eigenvalues of
the L matrix given by (6.71) are real; for complex conjugate eigenvalues, the
unstable disturbances are time periodic and one observes spatio-temporal
structures taking the form of propagating waves.
The existence of spatio-temporal patterns is not exclusive to fluid me-
chanics and chemistry. A multitude of self-organizations has been observed
in biology and living organisms, which are the most organized and complex
examples found in the nature. It has been conjectured that most of the prop-
erties of biological systems are the result of transitions induced by far from
equilibrium conditions and destabilizing mechanisms similar to autocatalytic
reactions. Because of their complexity, these topics will not be analysed here
but to further convince the reader about the universality of pattern forma-
tion, we prefer to discuss shortly three more examples of dissipative patterns
as observed in oceanography, electricity, and materials science.
6.6 Miscellaneous Examples of Pattern Formation
As recalled earlier, one observes many kinds of pattern formation in many
different systems. In this section, we give a concise overview of some of them.
6.6.1 Salt Fingers
In double diffusion convection, the flow instability is due to the coupling
of two diffusive processes, say heat and mass transport. In the case of the
170 6 Instabilities and Pattern Formation
salt fingers, which are of special interest in oceanography, the basic fields
are temperature and salinity, i.e. the concentration of salt in an aqueous
solution; in other situations, they are for example the mass concentrations of
two solutes in a ternary mixture at uniform temperature or the concentrations
of two polymers in polymeric solutions.
If we consider diffusion in a binary mixture submitted to a temperature
gradient, new features appear with respect to the simple Rayleigh–B´enard’s
convection. Apart from heat convection and mass diffusion, the Soret and
Dufour cross-effects are present and should be included in the expressions
of the constitutive equations, both for the heat flux q and the mass flow J.
Compared to the Rayleigh–B´enard’s problem, the usual balance equations of
mass, momentum, and energy must be complemented by a balance equation
for the mass concentration of one of the constituents while the mass density
is modified as follows to account for the mass concentration
ρ = ρ
0
[1 −α(T −T
0
) +β(c −c
0
)], (6.74)
where β stands for β = ρ
−1
0
(∂ρ/∂c). The equations for the disturbances are
then linearized in the same way as in Rayleigh–B´enard’s problem, and finite
amplitude solutions have also been analysed.
As a result of the very different values of the molecular D and heat χ
diffusivity coefficients (in salt sea waters D/χ = 10
−2
), some puzzling phe-
nomena are occurring. Instabilities arise when a layer of cold and pure water
is lying under a layer of hot and salty water with densities being such that
the cold water is less dense than the warm water above it; convection takes
then place in the form of thin fingers of up- and down-going fluids (Brenner
1970; McDougall and Turner 1982).
The mechanism responsible for the onset of instability is easily understood.
Imagine that, under the action of a disturbance, a particle from the lower
fresh cold water is moving upward. As heat conductivity is much larger than
diffusivity, the particle takes the temperature of its neighbouring but as it is
less dense than the saltier water outside it, the particle will rise upwards under
the action of an upward buoyancy force. Likewise if a particle from the hot
salt upper layer is sinking under the action of a perturbation, it will be quickly
cooled and, becoming denser than its surroundings, it creates a downward
buoyancy force accelerating the downward motion. This example illustrates
clearly the property that concentration non-homogeneities are dangerous for
the hydrodynamic stability of mixtures when their relaxation time is much
larger than that of temperature non-homogeneities.
The resulting finite amplitude motions have been called salt fingers be-
cause of their elongated structures (see Fig. 6.16). They have been observed
in a variety of laboratory experiments with heat-salt and sugar-salt mixtures
and in subtropical oceans.
6.6 Miscellaneous Examples of Pattern Formation 171
Fig. 6.16 Salt fingers (from Vidal et al. 1994)
Fig. 6.17 Temperature distributions in the ballast resistor
6.6.2 Patterns in Electricity
6.6.2.1 The Ballast Resistor
Let us first address some attention to the ballast resistor (Bedeaux et al.
1977; Pasmanter et al. 1978; Elmer 1992); it is an interesting example be-
cause it can be described by a one-dimensional model allowing for explicit
analytic treatments and, in addition, it presents useful technological aspects.
The device consists of an electrical wire traversing through a vessel of length
L filled with a gas at temperature T
G
. The control parameters are the tem-
perature T
G
and the electric current I crossing the wire. As much as the
temperature T
G
is lower than a critical value T
c
, the temperature of the wire
remains uniform but for T
G
> T
c
there is a bifurcation in the temperature
profile, which is no longer homogeneous, but instead is characterized by a
peak located at the middle of the electric wire (see Fig. 6.17).
It is worth to stress that, for T
G
> T
c
, the value of the electric current I
is insensitive to the variations of the electrical potential which indicates that
the ballast resistor can be used as a current stabilizer device.
172 6 Instabilities and Pattern Formation
6.6.2.2 The Laser
When the laser is pumped only weakly, one observes that the emitted light
waves have random phases with the result that the electric field strength
is a superposition of random waves. Above a critical pump strength, the
laser light becomes coherent (Haken 1977), meaning that the waves are now
ordered in a well-defined temporal organization (Fig. 6.18).
By increasing the external pumping, a sequence of more and more compli-
cated structures is displayed, just like in hydrodynamic and chemical insta-
bilities. In particular by pumping the laser strength above a second threshold,
the continuous wave emission is transformed into ultra-short pulses.
Fig. 6.18 Laser instability: the electrical field strength E is given as a function
of time: (a) disordered state, (b) ordered state, and (c) the same above the second
threshold
6.6.3 Dendritic Pattern Formation
Formation of dendrites, i.e. tree-like or snowflake-like structures as shown
in Fig. 6.19, is a much-investigated subject in the area of pattern formation.
Fig. 6.19 Dendritic xenon crystal growth (from Gollub and Langer 1999)
6.6 Miscellaneous Examples of Pattern Formation 173
Research on dendritic crystal growth has been motivated by the necessity to
better understand and control metallurgical microstructures.
The process which determines the formation of dendrites is essentially the
degree of undercooling that is the degree to which the liquid is colder than
its freezing temperature. The fundamental rate-controlling mechanism is dif-
fusion, either diffusion of latent heat away from the liquid–solid interface, or
diffusion of chemical species toward and away from this solidification front.
These diffusion processes lead to shape instabilities, which trigger the forma-
tion of patterns in solidification. In a typical sequence of events, the initially
crystalline seed immersed in its liquid phase grows out rapidly in a cascade of
branches whose tips move outwards at a given speed. These primary arms be-
come unstable against side branching and the new side branching are in turn
unstable with respect to further side branching, ending in a final complicated
dendritic structure. The speed at which the dendrites grow, the regularity,
and the distances between the side branches determine most of the prop-
erties of the solidified material, like its response to heating and mechanical
deformation.
To summarize, we have tried in this chapter to convince the reader
of the universality of pattern-forming phenomena. We have stressed that
similar patterns are observed in apparently very different systems, as il-
lustrated by examples drawn from hydrodynamics (Rayleigh–B´enard’s and
B´enard–Marangoni’s convections, Taylor’s vortices), chemistry (Belousov–
Zhabotinsky’s reaction and Turing’s instability), electricity (ballast resis-
tor and laser instability), and materials science (dendritic formation). Of
course, this list is far from being exhaustive and further applications have
been worked out in a great variety of areas. Figure 6.20 displays some ex-
amples like a quasi-crystalline standing-wave pattern produced by forcing a
layer of silicone oil at two frequencies (a), a standing-wave pattern of granular
material-forming stripes (b), a typical mammalian coat as the leopard (c).
Two last remarks are in form. That a great number of particles, of the
order of 10
23
, will behave in a coherent matter despite their random thermal
Fig. 6.20 Examples of patterns in quasi-crystalline pattern (a), granular material
(b), and typical leopard’s coat (c)
174 6 Instabilities and Pattern Formation
agitation is the main feature of pattern formation. As pointed out throughout
this chapter, self-organization finds its origin in two causes: non-linear dy-
namics and external non-equilibrium constraints. Fluctuations arising from
the great number of particles and their random motion are no longer damped
as in equilibrium but may be amplified with the effect to drive the system
towards more and more order. This occurs when the control parameter, like
the temperature gradient in B´enard’s experiment, crosses a critical point at
which the system undergoes a transition to a new state, characterized by
regular patterns in space and/or in time.
It may also be asked why appearance of order is not in contradiction
with the second law of thermodynamics which states that the universe is
evolving towards more and more disorder. There is of course no contradiction,
because the second principle, as enounced here, refers to an isolated system
while pattern forming can only occur in closed and/or open systems with
exchange of energy and matter with the surrounding. The decrease of entropy
in individual open or closed cells is therefore consistent with the entropy
increase of the total universe and the validity of the second law is not to be
questioned.
6.7 Problems
6.1. Non-linear Landau equation. Show that the solution of the non-linear
Landau equation
d
dt
[A[
2
= 2(Re σ) [A[
2
−2l [A[
4
is given by (6.12).
6.2. Landau equation and Rayleigh–B´enard’s instability. The Landau equa-
tion describing Rayleigh–B´enard’s instability can be cast in the form dA/dt =
σA − lA
3
whose steady solution is A
s
= (σ/l)
1/2
. Expanding σ around the
critical Rayleigh number, one has σ = α[Ra − (Ra)
c
], where α is a positive
constant from which follows the well-known result
A
s
= (α/l)
1/2
[Ra −(Ra
c
)]
1/2
.
Study the stability of this steady solution by superposing to it an infinites-
imally small disturbance A

and show that the steady non-linear solution is
stable for Ra > (Ra)
c
.
6.3. Third-order Landau equation. Consider the following third-order Landau
equation
dA
dt
= σA+αA
2
+βA
3
,
when A > 0, σ > 0, α < 0 and β
2
> 4ασ. For sufficiently small values of A
at t = 0, show that A tends to the equilibrium value A
e
= −σ/α +O(σ
2
) as
t →∞.
6.7 Problems 175
6.4. Rayleigh–B´enard’s instability. Consider an incompressible Boussinesq
fluid layer between two rigid horizontal plates of infinite extent. The two
plates are perfectly heat conducting and the fluid is heated from below. (a)
Establish the amplitude equations in the case of infinitesimally small distur-
bances (linear approximation). (b) Determine the marginal instability curve
Ra(k) between the dimensionless Rayleigh number Ra and wave number k.
(c) Calculate the critical values (Ra)
c
and k
c
corresponding to onset of con-
vection.
6.5. Rotating Rayleigh–B´enard’s problem. Two rigid horizontal plates extend-
ing to infinity bound a thin layer of fluid of thickness d. The system, subject
to gravity forces, is heated from below and is rotating around a vertical axis
with a constant angular velocity Ω. Determine the marginal curve Ra(k)
as a function of the dimensionless Taylor number Ta = 4Ω
2
d
4

2
, where ν
is the kinematic viscosity of the fluid. Does rotation play a stabilizing or a
destabilizing role?
6.6. Rayleigh–B´enard’s problem with a solute. Consider a two-constituent
mixture (solvent +solute) encapsulated between two free horizontal surfaces
and subject to a vertical temperature gradient β. Denoting by c(r, t) the
concentration of the solute, assume that the density of the mixture is given
by
ρ = ρ
0
[1 +α(T −T
0
) +γ(c −c
0
)].
Neglecting the diffusion of the solute so that dc/dt = 0, determine the mar-
ginal curve Ra(k) when the basic reference state is at rest with a given
concentration c
r
(z) and a temperature field T
r
= T
0
−βz.
6.7. Non-linear Rayleigh–B´enard’s instability. A thin incompressible fluid
layer of thickness d is bounded by two horizontal stress-free boundaries
(Ma = 0) of infinite horizontal extent. The latter are perfectly heat con-
ducting and gravity forces are acting on the fluid.
(a) Show that the convective motion can be described by the following non-
linear relation
∂w
∂t
+
1
2

3
w
2
∂z
3
= ∆
3
w −Ra ∆
1
w (0 < z < 1),
here w is the dimensionless vertical velocity component, Ra the Rayleigh
number and
∆ = ∆
1
+∂
2
/∂z
2
= ∂
2
/∂x
2
+∂
2
/∂y
2
+∂
2
/∂z
2
;
the corresponding boundary conditions (at z = 0 and z = 1) are
w =

2
w
∂z
2
=

4
w
∂z
4
= 0.
176 6 Instabilities and Pattern Formation
(b) Find the solution of the corresponding linear problem with normal mode
solutions of the form w = W(z)f(x, y) exp(σt); to be explicit, determine
the expressions of W(z) and σ and the equation satisfied by f(x, y).
(c) By assuming a non-linear solution of the roll type, i.e. w = Acos(kx)g(z),
show that the Landau equation associated to this problem can be written
as
dA
dt
= αA−βA
3
,
where α and β are two constants to be determined in terms of the data
Ra, (Ra)
c
, and k
c
.
6.8. Boundary condition with surface tension gradient. In dimensional form,
the kinematic boundary conditions at a horizontal surface normal to the z-
axis and subject to a surface tension gradient may be written as σ n = ∇S,
where σ is the stress tensor or more explicitly σ
xz
+ ∂S/∂x = 0 and σ
yz
+
∂S/∂y = 0. Show that, in non-dimensional form, the corresponding boundary
condition is given by (6.39) D
2
W = k
2
Maθ. Hint: After differentiating the
first relation with respect to x, the second with respect to y, make use of
Newton’s constitutive relation σ = η[∇v +(∇v)
T
] and the continuity relation
∇ v = 0.
6.9. B´enard–Marangoni’s instability. (a) Show that, in an incompressible liq-
uid layer whose lower boundary is in contact with a rigid plate while the upper
boundary is open to air and subject to a surface tension depending linearly
on the temperature, the marginal curve relating the Marangoni number Ma
to the wave number k is given by (6.40), i.e.
(Ma)
0
=
8k
2
cosh k(k −sinh k cosh k)
k
3
cosh k −sinh
3
k
,
both boundaries are assumed to be perfectly heat conducting and gravity
acceleration is neglected. (b) Find the corresponding critical values (Ma)
c
=
79.6 and k
c
= 1.99.
6.10. B´enard–Marangoni’s instability. The same problem as in 6.9, but now
with heat transfer at the upper surface governed by Newton’s cooling law
−λ
∂T
∂z
= h(T −T

),
where λ is the heat conductivity of the fluid, h the heat transfer coefficient,
and T

the temperature of the outside world, say the laboratory. In dimen-
sionless form, the previous law reads as DΘ = −Bi Θ, (D = d/dZ) with
Bi = hd/λ the so-called Biot number. The limiting case Bi = 0 corresponds
to an adiabatically isolated surface while Bi = ∞ describes a perfectly heat
conductor. (a) Determine the dependence of Ma with respect to k and Bi. (b)
Draw the marginal instability curves for Bi = 0, 1, 10. (c) Sketch the curves
(Ma)
c
(Bi) and k
c
(Bi).
6.7 Problems 177
6.11. The Rayleigh–B´enard–Marangoni’s instability. Show that, for the cou-
pled buoyancy–surface tension-driven instability, the Ra and Ma numbers
obey the relation (6.41)
Ra
(Ra)
c
+
Ma
(Ma)
c
= 1,
where (Ra)
c
is the critical Rayleigh number without Marangoni effect and
(Ma)
c
the critical Marangoni number in absence of gravity.
6.12. The Lorenz model. (a) Show that the steady solutions of (6.25) cor-
responding to supercritical convection are given (6.35) by A
1
= A
2
=
_
b(r −1) for r > 1. (b) Prove that this solution becomes unstable at
r = Pr(Pr b + 3)/(Pr − b − 1). (c) Solve numerically the Lorenz equations
for Pr = 10, b = 8/3, and r = 28 (Sparrow 1982).
6.13. Couette flow between two rotating cylinders. Consider a non-viscous
fluid contained between two coaxial rotating cylinders. The reference state
is stationary with u
r
= u
z
= 0, u
θ
(r) = rΩ(r), the quantity Ω(r) is an
arbitrary function of the distance r to the axis of rotation and is related to the
reference pressure by p
ref
= ρ
_
rΩ
2
(r)dr. (a) Show that the latter result is
directly obtained from the radial component of the momentum equation. (b)
Using the normal mode technique, show that, for axisymmetric disturbances
(∂/∂θ = 0), the amplitude equation is given by
(DD

−k
2
)U
r

k
2
σ
φ(r)U
r
= 0,
where D = d/dr, D

= D + 1/r, and φ(r) = (1/r
3
)d[(r
2
Ω)
2
]/dr is the so-
called Rayleigh discriminant. It is interesting to observe that the quantity
(r
2
Ω) in the Rayleigh discriminant is related to the circulation along a circle
of radius r by
_

0
u
θ
(r)r dθ = 2πr
2
Ω(r).
(c) Show further that the flow is stable with respect to axisymmetric dis-
turbances if φ ≥ 0. This result reflects the celebrated Rayleigh circulation
criterion stating that a necessary and sufficient condition of stability is that
the square of the circulation does not decrease anywhere.
6.14. Lotka chemical reactions. Show that the following sequence of chemical
autocatalytic reactions
A + X →2X
X + Y → 2Y
Y + B →E + D
where the concentrations of substances A and B are maintained fixed, corre-
spond to the Lotka–Volterra model.
Chapter 7
Extended Irreversible Thermodynamics
Thermodynamics of Fluxes: Memory
and Non-Local Effects
With this chapter, we begin a panoramic overview of non-equilibrium ther-
modynamic theories that go beyond the local equilibrium hypothesis, which
is the cornerstone of classical irreversible thermodynamics (CIT). We hope
that this presentation, covering Chaps. 7–11, will convince the reader that
non-equilibrium thermodynamics is a fully alive and modern field of research,
combining practical motivations and conceptual questions. Indeed, such basic
topics as the definition and meaning of temperature and entropy, the formu-
lation of the second law, and its consequences on the admissible transport
equations are still open questions nowadays.
Modern technology strives towards miniaturized devices and high-frequency
processes, whose length and timescales are comparable to the mean free path
of the particles and to the internal relaxation times of the devices, thus requir-
ing extensions of the classical transport laws studied in the previous chapters.
Indeed, these laws assume an instantaneous response of the fluxes to the im-
posed thermodynamic forces, whereas, actually, it takes some time for the
fluxes to reach the values predicted by the classical laws. As a consequence,
when working at short timescales or high frequencies, and correspondingly
at short length scales or short wavelengths, the generalized transport laws
must include memory and non-local effects. The analysis of these generalized
transport laws is one of the main topics in modern non-equilibrium ther-
modynamics, statistical mechanics, and engineering. Such transport laws are
generally not compatible with the local equilibrium hypothesis and a more
general thermodynamic framework must be looked for.
Going beyond CIT and exploring new frontiers are the driving impetus
for the development of recent non-equilibrium thermodynamic theories. In
that respect, we will successively analyse extended irreversible thermodynam-
ics (EIT), theories with internal variables, rational thermodynamics, Hamil-
tonian formulation, and mesoscopic theories.
This overview starts with EIT, because of its formal simplicity and its prox-
imity to the methods of CIT to which the reader is already acquainted. EIT
provides a macroscopic and causal description of non-equilibrium processes
179
180 7 Extended Irreversible Thermodynamics
and is based on conceptually new ideas, like the introduction of the fluxes as
additional non-equilibrium independent variables, and the search for general
transport laws taking the form of evolution equations for these fluxes. Such
equations will be generally obtained by considering the restrictions imposed
by the second law of thermodynamics.
To be explicit, in EIT the space 1 of state variables is formed by the union
of the space ( of classical variables like mass, momentum, energy, and com-
position, and the space T of the corresponding fluxes, i.e. 1 = ( ∪ T. The
physical nature of the T-variables is different from that of the (-variables.
The latter are slow and conserved with their behaviour governed by the clas-
sical balance laws. In contrast, the T-variables are fast and non-conserved:
they do not satisfy conservation laws and their rate of decay is generally very
short. In dilute gases, it is of the order of magnitude of the collision time
between molecules, i.e. 10
−12
s. This means that, for time intervals much
larger than this value, fast variables can be ignored. This is no longer true in
high-frequency phenomena, when the relaxation time of the fluxes is compa-
rable to the inverse of the frequency of the process, or in some materials, like
polymers, dielectrics, or superfluids, characterized by rather large relaxation
times of the order of seconds or minutes. The independent character of the
fluxes is also made evident when the mean free path of heat or charge carri-
ers becomes comparable to the dimensions of the sample, as in nano-systems.
Other motivations for elevating the fluxes to the rank of variables are given
at the end of this chapter.
The domain of application of EIT enlarges the frontiers of CIT, whose
range of validity is limited to small values of the relaxation times τ of the
fluxes, i.e. to small values of the Deborah number De ≡ τ/t
M
, with t
M
a
macroscopic timescale, and to small values of the Knudsen number Kn ≡ /L,
where is the mean free path and L a macroscopic length. The transport
equations derived from EIT reduce to the CIT expressions for De ¸ 1 and
Kn ¸ 1, but are applicable to describe a wider range of situations encom-
passing De > 1 and Kn > 1. Examples of situations for which De ≥ 1 are
processes where the macroscopic timescale becomes short enough to be com-
parable to the microscopic timescale, as in ultrasound propagation in dilute
gases or neutron scattering in liquids, or when the relaxation time becomes
long enough to be comparable to the macroscopic timescale, as in polymer
solutions, suspensions, superfluids, or superconductors. The property Kn ≥ 1
is characteristic of micro- and nano-systems as thin films, superlattices, sub-
micronic electronic devices, porous media, and shock waves.
A simple way, although not unique, to obtain the time evolution equations
of the fluxes on a macroscopic basis is to assume the existence of a generalized
entropy and to follow the same procedure as in CIT. In EIT, it is taken for
granted that there exists a non-equilibrium entropy s to which the following
properties are assigned:
• It is an additive quantity.
• It is a function of the whole set of variables s = s(1).
7.1 Heat Conduction 181
• It is a concave function of the state variables.
• Its rate of production is locally positive.
Once the expression of s is known, it is an easy matter to derive generalized
equations of state, which are of interest in the description of non-equilibrium
steady states.
The scope of this chapter is to give a general presentation of what EIT is,
how it works, and what can be expected from it. EIT has been the object of
several monographs by Jou et al. (2000, 2001), Lebon (1992) and M¨ uller and
Ruggeri (1998). For a more microscopic perspective, we refer to the books
by Eu (1992, 1998) and Luzzi et al. (2001, 2002); see also two collections of
contributions by several authors (Casas-V´ azquez et al. 1984; Sieniutycz and
Salamon 1992) or reviews by Garcia-Colin (1991, 1995) and Nettleton and
Sobolev (1995).
As an introductory example, we will study heat conduction in a rigid body
with memory effects; in this problem, only the heat flux is introduced as extra
variable. Afterwards, we shall discuss more complicated situations, such as
viscous fluids, polymer solutions, and electric transport in microelectronic
devices, where other fluxes, as the viscous pressure tensor and the electric
current, are selected as supplementary independent variables.
7.1 Heat Conduction
After outlining the shortcomings of the classical approach, we shall motivate
the choice of the heat flux as independent variable in the prototype problem
of heat conduction in a non-deformable solid at rest, and we will take advan-
tage of this particular example to introduce the main tenets of the general
formalism.
7.1.1 Fourier’s vs. Cattaneo’s Law
The best-known model for heat conduction is Fourier’s law, which relates
linearly the temperature gradient ∇T to the heat flux q according to
q = −λ∇T, (7.1)
where λ is the heat conductivity, depending generally on the temperature.
By substitution of (7.1) in the energy balance equation, written in absence
of source terms as
ρ
∂u
∂t
= −∇ q, (7.2)
and relating the specific internal energy u to the temperature by means of
du = c
v
dT, with c
v
being the heat capacity per unit mass at constant volume,
one obtains
ρc
v
∂T
∂t
= ∇ (λ∇T). (7.3)
182 7 Extended Irreversible Thermodynamics
The material time derivative has been replaced here by the partial time deriv-
ative because the body is at rest. From a mathematical point of view, (7.3)
is a parabolic differential equation. Although this equation is well tested for
most practical problems, it fails to describe the transient temperature field
in situations involving short times, high frequencies, and small wavelengths.
For example, it was shown by Maurer and Thomson (1973) that, by sub-
mitting a thin slab to an intense thermal shock, its surface temperature is
300

C larger than the value predicted by (7.3). The reasons for this failure
must be found in the physical statement of Fourier’s law, according to which
a sudden application of a temperature difference gives instantaneously rise to
a heat flux everywhere in the system. In other terms, any temperature dis-
turbance will propagate at infinite velocity. Physically, it is expected, and it
is experimentally observed, that a change in the temperature gradient should
be felt after some build-up or relaxation time, and that disturbances travel
at finite velocity. From a microscopic point of view, Fourier’s law is valid in
the collision-dominated regime, where there are many collisions amongst the
particles, but it loses its validity when one approaches the ballistic regime, in
which the dominant collisions are those of the particles with the boundaries
of the system rather than the collisions amongst particles themselves.
To eliminate these anomalies, Cattaneo (1948) proposed a damped version
of Fourier’s law by introducing a heat flux relaxation term, namely
τ
∂q
∂t
= −(q +λ∇T). (7.4)
The term containing the time τ represents the heat flux relaxation. When
the relaxation time τ of the heat flux is negligible or when the time variation
of the heat flux is slow, this equation reduces to Fourier’s law. For homoge-
neous solids, τ describes molecular-scale energy transfer by either phonons
or electrons, and it is very small, of the order of time between two successive
collisions at the microscopic level. Therefore, in most practical heat transfer
problems, infinite propagation is not relevant as those parts of the signals
with infinite velocity are strongly damped at room temperature. However,
when slow internal degrees of freedom are involved, as in polymers, superflu-
ids, porous media, or organic tissues, τ reflects the time required to transfer
energy between different degrees of freedom and it may be relatively large,
of the order or larger than 1 s.
Relaxational effects on heat conductors were already discussed by Maxwell
(1867), Cattaneo (1948), Vernotte (1958) and Grad (1958), but without refer-
ring to its thermodynamic implications, which will be analysed in Sect. 7.1.2
in the frame of EIT.
Assuming constant values of c
v
and λ and introducing (7.4) in (7.2) results
in the following hyperbolic equation
τ

2
T
∂t
2
+
∂T
∂t
−χ∇
2
T = 0, (7.5)
7.1 Heat Conduction 183
where χ = λ/ρc
v
designates the heat diffusivity. Equation (7.5) is sometimes
called telegrapher’s equation because it is similar to the one describing prop-
agation of electrical signals along a wire. For small values of the time t ¸τ,
the first term of (7.5) is dominant, so that it reduces to
τ

2
T
∂t
2
= χ∇
2
T. (7.6)
This is a wave equation with a wave propagating at the velocity (χ/τ)
1/2
; it
describes a reversible process, as it is invariant with respect to time inversion.
In contrast, for timescales much longer than τ(t ¸τ), the first term of (7.5)
is negligible and one obtains a partial differential equation of the form
∂T
∂t
= χ∇
2
T, (7.7)
which is associated with diffusion of heat, as shown in Chap. 2. Diffusion is
typically an irreversible process, as (7.7) is not invariant when t is changed
into −t. To summarize, at short times the transport equation (7.5) is re-
versible and heat propagates as a wave with a well-defined speed (which
may be microscopically interpreted as a ballistic motion of heat carriers),
whereas at longer times the process becomes irreversible and heat is diffused
throughout the system. It is therefore clear that τ can be interpreted as the
characteristic time for the crossover between ballistic motion and the onset of
diffusion. In the context of chaotic deterministic systems, τ is interpreted as
the Lyapunov time beyond which predictivity is lost (Nicolis and Prigogine
1989).
The dynamical properties of (7.5) have been thoroughly analysed. By
assuming that there exists a solution of the form T(x, t) = T
0
exp[i(kx−ωt)],
where ω is a (real) frequency and k is a (complex) wave number, it is found
(see Problem 7.2) that the solution is characterized by a phase speed v
p
and
an attenuation length α, respectively, given by
v
p
=
ω
Re k
=

2χω
_
τω +

1 +τ
2
ω
2
, α =
1
Imk
=

v
p
. (7.8)
In the low-frequency limit (τω ¸1), it is found that
v
p,0
=
_
2χω, α
0
=
_
2χ/ω, (7.9)
which are the results obtained directly from Fourier’s law. In the high-
frequency limit (τω ¸1), the phase speed and attenuation length are
v
p,∞
≡ U = (χ/τ)
1/2
, α

= 2(χτ)
1/2
. (7.10)
The velocity U corresponds to the so-called second sound, which is a tem-
perature wave, not to be confused with the first sound, which is a pressure
wave with velocity c
0
=
_
(∂p/∂ρ)
s
. The value of v
p,∞
diverges when the
184 7 Extended Irreversible Thermodynamics
relaxation time vanishes, thus leading to an infinite speed of propagation.
The presence of the relaxation time represents much more than a minor cor-
rection to the classical results, as in the high-frequency regime it leads to a
completely different behaviour than that predicted by the classical Fourier’s
law. In Sect. 7.1.3, we discuss other physical consequences of (7.5). The analy-
sis of thermal waves has been the topic of much research (Joseph and Preziosi
1989; Dreyer and Struchtrup 1993; Tzou 1997; Jou et al. 2001).
The differences between parabolic and hyperbolic equations are shown in
Box 7.1. Figure 7.1 provides a qualitative comparison between the parabolic
(τ = 0) and hyperbolic (τ ,= 0) solutions of (7.5), at a short time after appli-
cation of a thermal pulse in a one-dimensional rod. According to Cattaneo’s
Box 7.1 Differences Between Hyperbolic and Parabolic Heat
Transport
As shown in Box 2.3, the response to a delta pulse of temperature is given,
according to the classical diffusion equation, by
∆T(x, t) =
g
0
2(πχt)
1/2
exp
_

x
2
4χt
_
, (7.1.1)
where g
0
is a value which depends on the intensity of the initial pulse.
The response to a delta pulse predicted by the telegrapher’s equation is
much more complicated, but has been derived in many textbooks (see, in
particular, Morse and Feshbach 1953; Jou et al. 2001; Tzou 1997). It is
found that
∆T(x, t) =
g
0
4
exp
_

t

__
2I
0
(ξ)Uτδ(Ut −[x[)
+
_
I
0
(ξ) +
t
τξ
dI
0
(ξ)

_
H(Ut −[x[)
_
, (7.1.2)
where I
0
(x) is the modified Bessel function of the first kind, H(x) the
Heaviside function, U the second sound velocity given by U ≡ (χ/τ)
1/2
and ξ ≡
1
2
[(t/τ)
2
− (x/Uτ)
2
]
1/2
. The main differences with the solution
of the classical diffusion equation (7.1.1) are shown in Fig. 7.1. They are
essentially:
1. In the classical description, ∆T(x, t) is non-zero at any position in space
for t > 0. In contrast, ∆T(x, t) given by (7.1.2) vanishes for x > Ut, U
being the speed of the front.
2. In (7.1.2) there are two main contributions: the first, in δ(Ut − x), de-
scribes the propagation of the initial pulse with speed U, damped by the
exponential term. The second term, in H(Ut −x), is related to the wake,
which, for sufficiently long times Ut ¸x, tends to the diffusive solution
(7.1.1).
7.1 Heat Conduction 185
Fig. 7.1 Fourier (solid line) and Cattaneo (dashed line) temperature distributions
in one-dimensional rod: (a) corresponds to a given time t
1
and (b) corresponds to a
time t
2
> t
1
law, the wave propagates with a speed v
p,∞
= (χ/τ)
1/2
through the sys-
tem; at the right of the wave front, located at = v
p,∞
t = (χ/τ)
1/2
t, the
temperature perturbation is not felt. The quantity is called the penetra-
tion depth, which in Fourier’s law (τ = 0) extends to infinity, making that
the disturbance is instantaneously felt throughout the system (see Fig. 7.1a).
The temperature behind the wave front will be higher than the one corre-
sponding to Fourier’s law, because the same amount of energy is confined
into a smaller volume. When the time is increased, the diffusion of energy
over a larger volume has a damping effect on the wave, and the temperature
distribution becomes closer to the Fourier prediction (Fig. 7.1b). As time in-
creases, the difference between the hyperbolic and the parabolic temperature
distributions becomes smaller and smaller.
Despite the success of Cattaneo’s model and its generalizations to describe
heat transfer at high frequencies, its thermodynamic consequences are less
known and are worth examining. First, it should be stressed that Cattaneo’s
equation is not compatible with CIT. Indeed, after substitution of Cattaneo’s
equation (7.4) in the classical expression for the entropy production as derived
in Chap. 2, namely σ
s
= q ∇T
−1
, it is found that
σ
s
=
λ
T
2
(∇T)
2
+
τ
T
2
∂q
∂t
∇T, (7.11)
which is no longer positive definite, because of the presence of the second
term of the right-hand side. The problem finds its origin in the local equi-
librium hypothesis, which must therefore be revisited. This is better seen by
examining the time evolution of entropy in an isolated rigid body. Making
use of the local equilibrium assumption, one obtains for the total entropy of
the system a non-monotonic increase, while use of EIT yields a monotonic in-
crease as it is shown by the solid curve in Fig. 7.2 (Criado-Sancho and Llebot
1993).
186 7 Extended Irreversible Thermodynamics
Fig. 7.2 The evolution of the classical entropy S
CIT
during the equilibration of an
isolated system when use is made of Cattaneo’s equation is given by the dashed curve.
The evolution of the extended entropy S
EIT
, obtained from (7.2.15), is represented
by the solid curve, which, in contrast with that of S
CIT
, increases monotonically
Instead of being monotonically increasing, the classical entropy behaves in
an oscillatory way, as shown in Box 7.2. Strictly speaking, this result is not
incompatible with the Clausius’ formulation of the second law, which states
that the entropy of the final equilibrium state must be higher than the en-
tropy of the initial equilibrium state. However, the non-monotonic behaviour
of the entropy is in contradiction with the local equilibrium formulation of
the second law, which requires that the entropy production must be positive
everywhere at any time. Note, finally, that Cattaneo’s model does not pre-
clude that heat flows from lower to higher temperature. Nevertheless, this is
not in contradiction with the Clausius’ statement of the second law, about
the impossibility of constructing a cyclic heat engine whose only effect is
to transport heat from a colder to a hotter reservoir, because in (7.4) the
mentioned unusual behaviour lasts only over time intervals of the order of
the relaxation time during which Clausius’ claim is at least questionable.
Notwithstanding, this behaviour is in opposition with the local equilibrium
formulation of the second law, requiring that heat should always move from
hotter to colder regions, in absence of any other effect.
Box 7.2 Evolution of Entropy in a Discrete Isolated System: CIT
vs. EIT Behaviour
We study heat transfer between two rigid bodies at temperatures T
1
and T
2
(< T
1
) put into contact. Let the total system be isolated from the
outside world and let
˙
Q be the amount of heat exchanged between the
7.1 Heat Conduction 187
two bodies per unit time. In the local equilibrium formulation, one has
S(U
1
, U
2
) = S
1
(U
1
) +S
2
(U
2
) and therefore
dS
dt
=
dS
1
dt
+
dS
2
dt
= T
−1
1
dU
1
dt
+T
−1
2
dU
2
dt
. (7.2.1)
Since the global system is isolated, dU
1
+ dU
2
= 0, and from the first law
of thermodynamics, one has
dU
1
dt
= C
1
dT
1
dt
= −
˙
Q,
dU
2
dt
= C
2
dT
2
dt
=
˙
Q, (7.2.2)
where use has been made of the equations of state dU
i
= C
i
dT
i
, and C
i
are
the respective heat capacities of the systems.
For an isolated system, (7.2.1) represents the rate of entropy produced
inside the system, and in virtue of (7.2.2), it can be written as
dS
dt
= −(T
−1
1
−T
−1
2
)
˙
Q, (7.2.3)
or assuming that T
1
−T
2
≡ ε is small,
dS
dt
=
ε
T
2
˙
Q, (7.2.4)
where T is an intermediate temperature between T
1
and T
2
. According
to the second law of thermodynamics, (7.2.3) must be non-negative. This
implies that heat only flows from the region of highest temperature to the
region of lowest temperature, according to the original Clausius’ formulation
of the second law.
The simplest hypothesis ensuring the positiveness of (7.2.4) is to as-
sume, as in Fourier’s law, that the heat flux
˙
Q is proportional to the driving
force ε/T
2
, so that
˙
Q = K
ε
T
2
, (7.2.5)
with K being a positive coefficient. Combining (7.2.2) with (7.2.5) one
obtains for the evolution of ε

dt
= −C
−1
˙
Q, (7.2.6)
where C is defined as C
−1
= C
−1
1
+ C
−1
2
. When (7.2.5) is introduced into
(7.2.6), one finds that

dt
= −
K
CT
2
ε. (7.2.7)
Thus ε decays exponentially as ε = ε
0
exp[−Kt/(CT
2
)]. Now, substitution
of (7.2.4) in (7.2.3) leads to
dS
dt
= K
ε
2
T
2
≥ 0, (7.2.8)
188 7 Extended Irreversible Thermodynamics
from which follows that the entropy is a monotonically increasing function
of time.
Consider now the more general situation in which heat transfer is de-
scribed by the Cattaneo’s law
τ
d
˙
Q
dt
+
˙
Q = K
ε
T
2
, (7.2.9)
where τ is the relaxation time of
˙
Q. After combining (7.2.6) and (7.2.9),
one finds that the evolution of ε is governed by
τ
d
2
ε
dt
2
+

dt
+
K
CT
2
ε = 0, (7.2.10)
which is similar to the equation of motion of a damped pendulum. The
decay of ε is no longer exponential but will exhibit oscillatory behaviour
for 4Kτ/(CT
2
) > 1. The term in dε/dt corresponds to diffusion of heat,
whereas the term in d
2
ε/dt
2
describes the propagation of the heat wave.
Substituting the solution of (7.2.10) in (7.2.3) with
˙
Q given by (7.2.5) leads
to an expression for the entropy S which exhibits a non-monotonic depen-
dence with respect to time, as indicated by the dashed curve in Fig. 7.2.
In EIT, the quantity
˙
Q is viewed as an independent variable of a gen-
eralized entropy S
EIT
, whose rate of variation is, up to the second order in
˙
Q,
dS
EIT
dt
= T
−1
1
dU
1
dt
+T
−1
2
dU
2
dt
−a
˙
Q
d
˙
Q
dt
(7.2.11)
instead of (7.2.1). Using the conservation of energy (7.2.2), one may write
(7.2.11) as
dS
EIT
dt
= −
_
T
−1
1
−T
−1
2
_
˙
Q−a
˙
Q
d
˙
Q
dt
, (7.2.12)
which, for small temperature differences, simplifies to
dS
EIT
dt
=
_
ε
T
2
−a
d
˙
Q
dt
_
˙
Q. (7.2.13)
Assuming a linear relation between the flux
˙
Q and the force, given by the
quantity inside brackets, one gets
˙
Q = K
_
ε
T
2
−a
d
˙
Q
dt
_
(7.2.14)
wherein positiveness of the entropy production requires that K > 0. Making
use of (7.2.6) and replacing (7.2.14) by (7.2.13), one obtains
dS
EIT
dt
=
C
2
K
_

dt
_
2
. (7.2.15)
7.1 Heat Conduction 189
This expression is either positive or zero, but never negative. It is therefore
concluded that S
EIT
, whose evolution is plotted by the continuous curve in
Fig. 7.2, increases monotonically in the course of time and is thus compatible
with evolution equations of the Cattaneo type. For an explicit analysis of
the continuous problem, see Criado-Sancho and Llebot (1993).
7.1.2 Extended Entropy
Going back to Cattaneo’s equation (7.4), one observes that it represents truly
an evolution equation for the heat flux, which can therefore be considered as
an independent variable. This is indeed the key of EIT, where it is assumed
that the entropy s = s(u, q) does depend not only on the classical variable,
namely the specific internal energy u, but also on the heat flux q.
In differential form, the generalized entropy is written as follows:
ds =
_
∂s
∂u
_
du +
_
∂s
∂q
_
dq. (7.12)
In analogy with the classical theory, one can define a non-equilibrium tem-
perature θ(u, q
2
) by means of the reciprocal of (∂s/∂u)
q
. This possibility is
examined in Box 7.3. Here, we will neglect second- and higher-order contribu-
tions in q and simply identify ∂s/∂u with the inverse of the local equilibrium
temperature T(u), so that
∂s
∂u
=
1
T(u)
. (7.13)
The remaining partial derivative in (7.12) will be denoted as
_
∂s
∂q
_
= −T
−1

1
(u)q, (7.14)
wherein the minus sign and the factor T
−1
v are introduced for convenience,
with v the specific volume (namely, the reciprocal of the mass density); α
1
is a
undetermined function of u alone because, like temperature, its dependence
on q
2
is assumed to be negligible. Accordingly, the final expression of the
generalized Gibbs’ equation (7.12) is
ds = T
−1
du −T
−1

1
q dq. (7.15)
From this expression and the internal energy balance law (7.2), one obtains,
for the material time derivative ˙ s of the entropy,
ρ˙ s = −T
−1
∇ q −T
−1
α
1
q ˙ q, (7.16)
190 7 Extended Irreversible Thermodynamics
or, equivalently,
ρ˙ s = −∇ (T
−1
q) +q (∇T
−1
−T
−1
α
1
q ˙ q). (7.17)
This equation can be cast in the general form of a balance equation
ρ˙ s = −∇ J
s

s
, (7.18)
with the entropy flux J
s
and the (positive) entropy production σ
s
given,
respectively, by
J
s
= T
−1
q (7.19)
and
σ
s
= q (∇T
−1
−T
−1
α
1
˙ q) ≥ 0. (7.20)
Relation (7.20) has the structure of a bilinear form σ
s
= q X in the flux
q and the force X, identified as the quantity within parentheses in (7.20);
it differs from the classical thermodynamic force, which is simply ∇T
−1
, by
the presence of a term in the time derivative of the heat flux. The simplest
way to obtain an evolution equation for q compatible with the positiveness
of σ
s
is to assume that the force X is linear in q, namely
∇T
−1
−T
−1
α
1
˙ q = µ
1
q, (7.21)
where the phenomenological coefficient µ
1
may depend on u but not on q
because, as previously, third-order contributions in q are omitted. Introduc-
tion of (7.21) into (7.20) results in σ
s
= µ
1
q q ≥ 0, from which is inferred
that µ
1
> 0.
Expression (7.21) contains two non-defined coefficients α
1
and µ
1
, which
must be identified on physical grounds. Under steady state conditions, (7.21)
simplifies to
q = −
1
µ
1
T
2
∇T. (7.22)
A comparison with Fourier’s law q = −λ∇T yields then µ
1
= (λT
2
)
−1
from
which it results λ ≥ 0. Next, by comparing (7.21) with Cattaneo’s equation
(7.4), one is led to
α
1
= τ/λT. (7.23)
Box 7.3 Non-Equilibrium Temperature vs. Local Equilibrium
Temperature
In analogy with the classical theory, we define the non-equilibrium temper-
ature θ by
θ
−1
(u, q) =
_
∂s
∂u
_
q
. (7.3.1)
Expanding θ
−1
around q = 0, and omitting terms of order higher than q
2
,
one can write
θ
−1
(u, q) = T
−1
(u) +α(u)q
2
, (7.3.2)
7.1 Heat Conduction 191
where T designates the local equilibrium temperature; the coefficient α(u)
depends only on u at this order of approximation.
With the identification (7.23), the generalized Gibbs’ equation (7.15)
takes the form
ds = θ
−1
du −
τ
ρλT
2
q dq. (7.3.3)
The integrability condition of (7.3.3) demands that
∂θ
−1
∂q
= −

∂u
_
τ
ρλT
2
_
q, (7.3.4)
and, after integration,
θ
−1
(u, q) = T
−1
(u) −
1
2

∂u
_
τ
ρλT
2
_
q q. (7.3.5)
This result is interesting because it provides an explicit expression of the
lowest order correction of the temperature in systems out of equilibrium.
Note that θ can be identified with T when the quantity τ/ρλT
2
is con-
stant. For an overview of the current discussions on the meaning and conse-
quences of temperature in non-equilibrium systems, see Casas-V´azquez and
Jou (2003).
With the above identifications of µ
1
and α
1
, the generalized Gibbs’ equa-
tion (7.15) takes the form
ds = T
−1
du −
τ
ρλT
2
q dq, (7.24)
wherein it is important to observe that the coefficient of the new term in dq
is completely identified in terms of physical quantities, namely the relaxation
time τ and the heat conductivity λ. After integration of (7.24), the explicit
expression for the entropy outside (local) equilibrium up to second-order
terms in q is
ρs(u, q) = ρs
eq
(u) −
1
2
τ
λT
2
q q. (7.25)
The monotonic increase of extended entropy as shown in Fig. 7.2 is due to
the last term on the right-hand side of (7.25).
7.1.3 Non-Local Terms: From Collision-Dominated
Regime to Ballistic Regime
However, all is not well with the Cattaneo’s equation. Although it is quali-
tatively satisfactory, as it predicts that heat pulses and high-frequency ther-
mal waves will propagate at finite speed, some quantitativepredictions are
192 7 Extended Irreversible Thermodynamics
not sufficiently accurate. For instance, the theory of solids predicts that
χ/τ = c
2
0
/3 in the so-called Debye approximation, with c
0
the sound velocity.
After introducing this value in (7.10), it is found that the second sound is
given by the constant value U = c
0
/

3, in contradiction with experiments
showing that U depends on temperature. Similarly, the kinetic theory of gases
predicts that, for monatomic gases of mass m, χ/τ = 5k
B
T/3m from which it
follows that U =
_
5k
B
T/3m, but this value deviates by more than 20% from
experimental data. One may then ask how to make EIT compatible with the
above experimental evidences. It appeared rather soon that the above dis-
crepancies find their origin in non-local effects, which were not incorporated
in Cattaneo’s original equation.
Non-local effects become important when the mean free path of the heat
carriers becomes comparable to the wavelength of the external perturba-
tion or to the dimensions L of the system. In this case, the Knudsen number
Kn = /L is of the order or larger than 1 and the transport laws change from
a collision-dominated regime, i.e. a regime with many collisions amongst the
particles, to a ballistic regime, where there are few collisions amongst the par-
ticles and the predominant interactions are collisions of the particles with the
walls. The difference of behaviour of transport in both situations is remark-
able. For instance, in heat pulse experiments, one distinguishes several modes
of transport: conduction, thermal waves, and ballistic fronts, and therefore
transport equations combining these factors are desirable.
The question then arises how to include these effects in the formalism.
A way out is to introduce a new extra variable, the flux of the heat flux,
described by a second-order tensor Q and to write, instead of the Cattaneo’s
equation (7.4), the following expression
τ
1
∂q
∂t
= −(q +λ∇T) +∇ Q. (7.26)
The tensor Q, assumed to be symmetric as confirmed by the kinetic theory
of phonons (Dreyer and Struchtrup 1993), may be split in the usual form
Q = QI +
0
Q, the scalar Q being one-third of its trace and
0
Q the deviatoric
part. In a relaxational approach, the evolution equations for Q and
0
Q may
be written as
τ
0
∂Q
∂t
= −Q+β

∇ q, (7.27)
τ
2

0
Q
∂t
= −
0
Q+ 2β

(∇
0
q)
s
, (7.28)
where superscript s refers to the symmetric part of the tensor. Note that,
when the relaxation times are neglected, these expressions parallel the
Newton–Stokes’ equations, with q playing the role of the fluid velocity and Q
7.1 Heat Conduction 193
that of the viscous pressure tensor. This is characteristic of the hydrodynamic
regime of phonon’s flow, which is indeed described by these equations.
Assuming that the relaxation times τ
0
and τ
2
are negligibly small, substi-
tuting (7.27) and (7.28) into (7.26) and restricting the analysis to the linear
approximation, one obtains, for the evolution equation of the heat flux,
τ
1
∂q
∂t
= −(q +λ∇T) +β


2
q +
_
β

+
1
3
β

_
∇(∇ q). (7.29)
Expression (7.29) is comparable with that obtained by Guyer and Krumhansl
(1966) from phonon kinetic theory. In this approach, heat transport is the
result of energy and momentum exchanges between massless colliding carri-
ers, called phonons. To be more explicit, the phonons undergo two types of
collisions: resistive R collisions which are collisions with defects of the lattice
and/or the boundaries of the crystal, and so-called Umklapp phonon–phonon
processes; they conserve energy but not momentum and have a characteristic
time τ
R
. The second type of collisions are referred to as normal N processes;
they conserve energy and momentum, and their characteristic time will be de-
noted by τ
N
. Starting from the Boltzmann’s equation, Guyer and Krumhansl
were able to derive the following equation for the heat flux:
∂q
∂t
+
1
τ
R
q +
1
3
ρc
v
c
2
0
∇T =
1
5
c
2
0
τ
N
_

2
q + 2∇(∇ q)
¸
. (7.30)
Comparison between (7.29) and (7.30) yields the following identifications
τ
1
= τ
R
,
λ
τ
R
=
1
3
ρc
v
c
2
0
, β

=
1
5
c
2
0
τ
N
τ
R
, β

=
1
3
c
2
0
τ
N
τ
R
, (7.31)
from which follows that τ
1
is associated with resistive collisions while the
non-local contributions find their origin in the normal collisions. The result
(7.29) contains as particular cases the Fourier’s and the Cattaneo’s equations
for heat transport. Indeed:
1. For 1/τ
R
and 1/τ
N
→∞, which corresponds to very high frequency of R
and N collisions, one recovers Fourier’s law.
2. For 1/τ
R
finite, 1/τ
N
→ ∞, the frequency of normal N phonon–phonon
collisions is large compared to that of R collisions and Cattaneo’s equation
is well suited.
3. In the general case where the frequencies of R and N collisions are com-
parable, non-local effects are important and Guyer–Krumhansl equation
is more adequate. It becomes also clear that the flux of the heat flux Q ac-
counts for the presence of the momentum-preserving N phonon–phonon
collisions.
Combining (7.29) with the energy balance equation and formulating the
final result in the Fourier’s space (i.e. in terms of frequency ω and wave
number k), the dispersion relation between ω and k for thermal waves is (see
Problem 7.5)
194 7 Extended Irreversible Thermodynamics
iω =
−χk
2
1 + iωτ
R
+
l
2
k
2
1+iωτ
N
, (7.32)
with l
2
=
4
3
β

for longitudinal waves and l
2
= β

for transverse waves.
For ωτ
R
¸ 1 ¸ ωτ
N
and ignoring non-local effects ( = 0), (7.32) leads to
the phase velocity v
p
= (χ/τ
R
)
1/2
= c
0
/

3. This is the regime referred to
as second sound, and is attained when τ
N
and τ
R
are of different orders of
magnitude. At higher frequencies, when both ωτ
N
¸ 1 and ωτ
R
¸ 1, the
phase velocity obtained from (7.32) for longitudinal waves is (Dreyer and
Struchtrup 1993, Dedeurwaerdere et al. 1994)
v
p
=
_
χ
τ
R
+
l
2
τ
R
τ
N
_
1/2
=
_
14
15
c
0
. (7.33)
This regime is typical of ballistic propagation of phonons, i.e. a flow in which
phonons travel through the whole crystal without suffering any collision. In
fact, the theoretical result (7.33) is not fully satisfactory because experiments
assign to the ballistic speed, a value equal to c
0
.
Better agreement with experiments may be achieved when higher-order
fluxes Q
1
, Q
2
, . . . , Q
N
are incorporated as independent variables in the for-
malism. Indeed, in some situations, as in diluted gases, the relaxation times
of the different higher-order fluxes are of the same order as that of the heat
flux itself, in such a way that when q is taken as an independent variable,
all higher-order fluxes should also be taken into account. In this way, one ob-
tains for the different fluxes a hierarchy of evolution equations for the infinite
number of higher-order fluxes of the form
τ
n
˙
Q
n
= −Q
n
+
l
2
n−1
τ
n−1
∇Q
n−1
−∇ Q
n+1
. (7.34)
Writing (7.34) in Fourier’s space ω k and solving the corresponding expres-
sion for the heat flux, one obtains a generalized Fourier’s law with a (ω, k)-
dependent thermal conductivity
˜ q(ω, k) = −ikλ(ω, k)
˜
T(ω, k), (7.35)
where ˜ q(ω, k) and
˜
T(ω, k) are the Fourier transforms of q(r, t) and T(r, t),
respectively. It is found that the generalized heat conductivity λ(ω, k) is given
by (Dedeurwaerdere et al. 1996; Jou et al. 2001)
λ(ω, k) =
λ(T)
1 + iωτ
1
+
k
2
l
2
1
1+iωτ
2
+
k
2
l
2
2
1+iωτ
3
+
k
2
l
2
3
1+iωτ
4
+···
, (7.36)
where l
n
are characteristic lengths of the order of the mean free path, and
τ
1
, τ
2
, . . . τ
N
are the relaxation times of the respective higher-order fluxes.
7.1 Heat Conduction 195
This result will be exploited in Sect. 7.1.4. Two types of truncations of the
continued-fraction expansion (7.36) have been considered. The most usual is
to assume that l
i
= 0 with i > n, for some given n. But a better approach
is to suppose that l
i
= l
n
for i > n. Asymptotic developments have been
performed, with the advantage that the whole formalism can be expressed
in terms of one single “effective” relaxation time, which incorporates in a
compact way the effect resulting from the introduction of an infinite number
of higher-order fluxes (see Jou et al. 2001).
7.1.4 Application to Steady Heat Transport
in Nano-Systems
We have mentioned that in nano-systems or submicronic electronic de-
vices, whose dimensions are comparable to the mean free path of phonons,
ballistic transport is dominant and, consequently, the validity of Fourier’s
law is questioned. Here we propose to revisit this law when the Knudsen
number Kn ¸ 1. To formulate the problem in simple terms, consider a
one-dimensional system of length L, whose opposite boundaries are at tem-
perature T and T +∆T, and transient effects are ignored. Depending on the
values of Kn, the heat flux q takes the following limiting forms
q = λ
∆T
L
(Kn ¸1, diffusive transport), (7.37a)
q = Λ∆T (Kn ¸1, ballistic transport), (7.37b)
where λ denotes the thermal conductivity and Λ is a heat conduction trans-
port coefficient. In the diffusive limit, the heat flux is proportional to the
temperature gradient, according to Fourier’s law; in contrast, in the ballistic
regime, the heat flux depends only on the temperature difference, but not
on the length L of the system. The values of λ and Λ have been derived
in the kinetic theory. For a monatomic ideal gas in the diffusive regime, it
is found that λ =
5
2
nk
B
(k
B
T/m)
1/2
, with n the particle number density,
m the mass of the particles, k
B
the Boltzmann constant, and the mean
free path given by = (k
B
T/m)
1/2
τ, with τ the average time between suc-
cessive collisions; in the rarefied gas regime, the heat conduction coefficient
Λ is given by Λ =
1
2
nk
B
(k
B
T/m)
1/2
. Note that the thermal conductivity
λ is proportional to , whereas Λ does not depend on it. It is also worth
to mention that computer simulations of heat transport in one-dimensional
systems suggest that in some situations Fourier’s law must be generalized in
the form q = λ(T)∆T/L
α
, with α an exponent whose value depends on the
details of the system. For instance, it is found that α = 0.63 for some an-
harmonic chains or one-dimensional gases, α = 0.5 for disordered harmonic
chains with free boundaries, or α = 1.5 for disordered harmonic chains with
fixed boundaries (Lepri et al. 2003).
196 7 Extended Irreversible Thermodynamics
A simple phenomenological modelling of the transition between the diffu-
sive and the ballistic regimes can be achieved by introducing a heat conduc-
tivity λ(T, /L), in such a way that in whole generality
q = λ(T, /L)
∆T
L
. (7.38)
The limiting values of this generalized conductivity should be
λ(T, /L) →λ(T) for /L →0,
λ(T, /L) →
λ(T)
a
L

≡ Λ(T)L for /L →∞,
where a is a constant depending on the system.
This may be achieved, for instance, through the continued-fraction ap-
proach (7.36), which in the steady state (ω
.
= 0), allows us to define a k-
dependent thermal conductivity λ(T, k). Since the system is characterized by
one single length scale, the length L, it seems natural to identify k with 2π/L.
This yields an expression for λ(T, /L), which in the simplest case where all
the l
n
take the same value, for instance, l
2
n
=
1
4

2
, and using the asymptotic
expression corresponding to (7.36) (Problem 7.6) yields
λ(T, /L) =
λ(T)L
2

2

2
_
_
1 + 4(π/L)
2
_
1/2
−1
_
, (7.39)
(see Jou et al. 2005). It is directly checked that (7.39) reduces to λ(T) for
/L → 0 and to λ(T)(L/) for /L → ∞; these are the required asymp-
totic behaviours. These considerations confirm that EIT provides a consis-
tent modelling for heat transfer processes taking place not only at short times
but also at micro- and nano-scales. Despite its simplicity, (7.39) satisfactorily
fits experimental results in silicon thin layers and nano-wires (Alvarez and
Jou 2007). Another possibility, which has been explored recently (Chen 2004;
Lebon et al. 2006b), is to split the heat flux into a diffusive and a ballistic
component, the latter being associated with fast particles and a long mean
free path, the former with slow particles and a short mean free path.
7.2 One-Component Viscous Heat Conducting Fluids
Up to now, we have only considered heat transfer in rigid bodies at rest. Let
us now proceed to the more general case of a compressible one-component
isotropic viscous fluid in motion subject to a temperature gradient. According
to EIT, the space of the thermodynamic state variables is the union of the
7.2 One-Component Viscous Heat Conducting Fluids 197
classical one (internal energy u and specific volume v) and the space of the
fluxes, namely the heat flux q, the bulk viscous pressure p
v
, and the shear
viscous pressure
0
P
v
. The generalized Gibbs’ equation takes the form
ds = T
−1
du +T
−1
p dv −T
−1

0
p
v
dp
v
−T
−1

1
q dq −T
−1

2
0
P
v
: d
0
P
v
,
(7.40)
where we have approximated the non-equilibrium temperature by the local
equilibrium temperature (see Box 7.3), similarly the non-equilibrium pressure
has been identified with the equilibrium pressure p, the coefficients α
0
.
, α
1
,
and α
2
are unknown scalar functions of u and v, their dependence with respect
to the fluxes is assumed to be negligible. Compared to CIT, the last three
terms are new. Our objective is to determine the evolution equations for the
fluxes satisfying the thermodynamic restrictions imposed by the second law.
We shall proceed in parallel with the procedure followed in Chap. 2, which
implies that we first determine the expression of the entropy production σ
s
.
To this end, we combine (7.40) with the expressions of ˙ u and ˙ v as derived
from the balance laws of mass (2.16) and energy (2.18). Omitting the energy
source term, it is easily found that
ρ˙ s = −T
−1
∇ q −T
−1
p
v
∇ v −T
−1
0
P
v
:
0
V−T
−1
α
0
p
v
˙ p
v
−T
−1
α
1
q ˙ q −T
−1
α
2
0
P
v
: (
0
P
v
)

. (7.41)
Before deriving the entropy production
σ
s
= ρ˙ s +∇ J
s
, (7.42)
we need an expression for the entropy flux J
s
. For isotropic systems, the
most general vector depending on u, v, q,
0
P
v
, and p
v
is, up to second order
in the fluxes,
J
s
= T
−1
q +β

p
v
q +β

0
P
v
q. (7.43)
We assume for simplicity that the coefficients β

and β

are constant; (7.43)
is supported by the kinetic theory of gases and may be viewed as a gener-
alization of the classical result J
s
= q/T. The entropy production is easily
derived from (7.42) by replacing ρ˙ s and J
s
, respectively, by their expressions
in (7.41) and (7.43). The final result is
σ
s
= q (∇T
−1


0
P
v

∇p
v
−T
−1
α
1
˙ q)
+p
v
(−T
−1
∇ v −T
−1
α
0
˙ p
v

∇ q)
+
0
P
v
: [−T
−1
0
V−T
−1
α
2
(
0
P
v
)

(∇
0
q)
s
]. (7.44)
198 7 Extended Irreversible Thermodynamics
Note that (7.44) has the structure of a bilinear form
σ
s
= q X
1
+p
v
X
0
+
0
P
v
:
0
X2
≥ 0, (7.45)
consisting of a sum of products of the fluxes q, p
v
, and
0
P
v
and their conjugate
generalized forces X
1
, X
0
, and
0
X2
. The latter are the respective terms be-
tween the brackets. They are similar to the expressions obtained in CIT, but
they contain additional terms depending on the time and space derivatives
of the fluxes.
The simplest flux–force relations ensuring the positiveness of σ
s
are
X
1
= µ
1
q, X
0
= µ
0
p
v
,
0
X2
= µ
2
0
P
v
, (7.46)
with µ
0
≥ 0, µ
1
≥ 0, and µ
2
≥ 0. Writing explicitly the expressions for the
generalized forces, one obtains in the linear approximation the following set
of evolution equations for the fluxes:
∇T
−1
−T
−1
α
1
˙ q = µ
1
q −β


0
P
v
−β

∇p
v
, (7.47)
−T
−1
∇ v −T
−1
α
0
˙ p
v
= µ
0
p
v
−β

∇ q, (7.48)
−T
−1
0
V−T
−1
α
2
(
0
P
v
)

= µ
2
0
P
v
−β

(∇
0
q)
s
. (7.49)
Note that in (7.49) (
0
P
v
)

stands for the time derivative of
0
P
v
.
The main features issued from the above thermodynamic formalism are:
• The positiveness of the coefficients µ
0
≥ 0, µ
1
≥ 0, and µ
2
≥ 0.
• The equality of the cross-coefficients relates q with ∇
0
P
v
and
0
P
v
with
(∇
0
q)
s
on the one side, q with ∇p
v
and p
v
with ∇q on the other side. This
is confirmed by the kinetic theory, and belongs to a class of higher-order
Onsager’s relations.
• The coefficients β

and β

appearing in the second-order terms of the
entropy flux (7.43) are the same as the coefficients of the cross-terms
in the evolution equations (7.47)–(7.49). This result is also in agree-
ment with the kinetic theory of gases. These coefficients describe the cou-
pling between thermal and mechanical effects; setting β

= β

= 0 means
absence of coupling in (7.47)–(7.49) and similarly in (7.43) of J
s
, which
boils down to the classical result J
s
= q/T.
To identify the coefficients appearing in (7.47)–(7.49) in physical terms,
consider the particular case where the space derivatives of the fluxes are
negligible. Equations (7.47)–(7.49) then reduce to
∇T
−1
−T
−1
α
1
˙ q = (λT
2
)
−1
q, (7.50)
−T
−1
∇ v −T
−1
α
0
˙ p
v
= (ζT)
−1
p
v
, (7.51)
−T
−1
0
V−T
−1
α
2
(
0
P
v
)
·
= (2ηT)
−1
0
P
v
. (7.52)
7.2 One-Component Viscous Heat Conducting Fluids 199
By setting
α
1
= τ
1
(λT)
−1
, α
0
= τ
0
ζ
−1
, α
2
= τ
2
(2η)
−1
, (7.53)
µ
1
= (λT
2
)
−1
, µ
0
= (ζT)
−1
, µ
2
= (2ηT)
−1
, (7.54)
(7.50)–(7.52) can be identified with the so-called Maxwell–Cattaneo’s laws
τ
1
˙ q +q = −λ∇T, (7.55)
τ
0
˙ p
v
+p
v
= −ζ∇ v, (7.56)
τ
2
(
0
P
v
)

+
0
P
v
= −2η
0
V, (7.57)
with λ > 0, ζ > 0, and η > 0 being the positive thermal conductivity, bulk
viscosity, and shear viscosity, respectively, and where τ
1
, τ
0
, and τ
2
are the
relaxation times of the respective fluxes. In terms of λ, ζ, η, and the relaxation
times τ
1
, τ
0
, and τ
2
, the linearized evolution equations (7.50)–(7.52) take the
following form:
τ
1
˙ q = −(q +λ∇T) +β

λT
2

0
P
v

λT
2
∇p
v
, (7.58)
τ
0
˙ p
v
= −(p
v
+ζ∇ v) +β

ζT∇ q, (7.59)
τ
2
(
0
P
v
)

= −(
0
P
v
+ 2η
0
V) + 2β

ηT(∇
0
q)
s
. (7.60)
Extension of the present description to the problem of thermodiffusion in
multi-component fluids has been achieved by Lebon et al. (2003). Despite in
ordinary liquids, the relaxation times are small, of the order of 10
−12
s, their
effects may be observed in neutron scattering experiments; in contrast, in
dilute gases or polymer solutions, the relaxation times may be rather large
and directly perceptible in light scattering experiments and in ultrasound
propagation. To illustrate the above analysis, let us mention that Carrassi and
Morro (1972) studied the problem of ultrasound propagation in monatomic
gases and compared the results provided by Maxwell–Cattaneo’s equations
(7.55)–(7.57) and the classical Fourier–Newton–Stokes’ laws.
In Table 7.1, the numerical values of the ultrasound phase velocities c
0
/v
p
(c
0
designating the sound speed) vs. the non-dimensional mean free path
of the particles are reported. It is observed that the classical theory devi-
ates appreciably from the experimental results as is increased, whereas the
Maxwell–Cattaneo’s theory agrees fairly well with the experimental data.
Table 7.1 Numerical values of c
0
/v
p
as a function of (Carrassi and Morro 1972)
0.25 0.50 1.00 2.00 4.00 7.00
(c
0
/v
p
)
(Fourier–Newton–Stokes)
0.40 0.26 0.19 0.13 0.10 0.07
(c
0
/v
p
)
(Maxwell–Cattaneo)
0.52 0.43 0.44 0.47 0.48 0.49
(c
0
/v
p
)
(experimental)
0.51 0.46 0.50 0.46 0.46 0.46
200 7 Extended Irreversible Thermodynamics
Taking into account the identifications (7.53), the explicit expression of
the entropy, after integration of (7.40), is
s
EIT
= s
eq
(u, v) −
τ
1
v
2λT
2
q q −
τ
0
v
2ζT
p
v
p
v

τ
2
v
4ηT
0
P
v
:
0
P
v
. (7.61)
This expression reduces to the local equilibrium entropy s
eq
(u, v) for zero
values of the fluxes. Since the entropy must be a maximum at equilibrium,
it follows that the relaxation times must be positive. Indeed, stability of
equilibrium demands that entropy be a concave function, which implies that
the second-order derivatives of s
EIT
with respect to its state variables are
negative; in particular,

2
s
EIT
∂q ∂q
= −

1
λT
2
< 0,

2
s
EIT

0
P
v
: ∂
0
P
v
= −

2
2ηT
2
< 0,

2
s
EIT
(∂p
v
)
2
= −

0
ζT
2
< 0.
(7.62)
Since the transport coefficients λ, ζ, and η must be positive as a consequence
of the second law, it turns out from (7.62) that the relaxation times are pos-
itive, for stability reasons. The property of concavity of entropy is equivalent
to the requirement that the field equations constitute a hyperbolic set. Hy-
perbolicity of evolution equations is characteristic of EIT and it is sometimes
imposed from the start as in Rational Extended Thermodynamics (M¨ uller
and Ruggeri 1998).
7.3 Rheological Fluids
Section 7.2 is dedicated to general considerations about the EIT description
of viscous fluid flows. As mentioned above, for most ordinary fluids, the re-
laxation times are generally very small so that for main problems the role of
relaxation effects is minute and can be omitted. However, this is no longer true
with rheological fluids, like polymer solutions, because the relaxation times
of the macromolecules are much longer than for small molecules, and they
may be of the order of 1 s and even larger. The study of rheological fluids has
been the concern of several thermodynamic approaches like rational ther-
modynamics, internal variables theories, and Hamiltonian formalisms (see
Chaps. 8–10). The main difference between EIT and other theories is that
in the former the shear viscous pressure is selected as variable, whereas in
other theories, variables related with the internal structure of the fluid, as
for instance, the so-called configuration tensor, are preferred. The EIT vari-
ables are especially useful in macroscopic analyses while internal variables
are generally more suitable for a microscopic understanding.
In the simple Maxwell model, which is a particular case of (7.60), the
viscous pressure tensor obeys the evolution equation
dP
v
dt
= −
1
τ
P
v
−2
η
τ
V. (7.63)
7.3 Rheological Fluids 201
This model captures the essential idea of viscoelastic models: the response to
slow perturbations is that of an ordinary Newtonian viscous fluid, whereas
for fast perturbations, with a characteristic time t of the order of the re-
laxation time τ or less, it behaves as an elastic solid. However, the mater-
ial time derivative introduced in (7.63) is not very satisfactory, neither for
practical predictions nor from a theoretical viewpoint, because it does not
satisfy the axiom of frame-indifference (see Chap. 9). This has motivated to
replace (7.63) by the so-called upper-convected Maxwell model
dP
v
dt
−(∇v)
T
P
v
−P
v
(∇v) = −
1
τ
P
v
−2
η
τ
V. (7.64)
In a steady pure shear flow corresponding to velocity components v =
( ˙ γy, 0, 0), with ˙ γ the shear rate, the upper-convected Maxwell model (7.64)
reads as
P
v
=


−2τη ˙ γ
2
−η ˙ γ 0
−η ˙ γ 0 0
0 0 0


. (7.65)
In contrast with the original Maxwell model, element P
v
11
of tensor (7.65)
contains a non-vanishing contribution, corresponding to the so-called normal
stresses. The upper-convected model agrees rather satisfactorily with a wide
variety of experimental data.
In (7.63) and (7.64), we have introduced one single relaxation time but
in many cases it is much more realistic to consider P
v
as a sum of several
independent contributions, i.e. P
v
=

j
P
v
j
with each P
v
j
obeying a linear
evolution equation such as (7.63) or (7.64), characterized by its own viscosity
η
j
and relaxation time τ
j
. These independent contributions arise from the
different internal degrees of freedom of the macromolecules. In this case, the
“extended” entropy should be written as
s(u, v, P
v
i
) = s
eq
(u, v) −
v
4T

i
τ
i
η
i
P
v
i
: P
v
i
(7.66)
and the corresponding model is known as the generalized Maxwell model.
In the above descriptions, the viscosity was supposed to be independent of
the shear rate. However, there exists a wide class of so-called non-Newtonian
fluids characterized by shear rate-dependent viscometric functions, like the
viscous coefficients. Such a topic is treated at full length in specialized works
on rheology and will not be discussed here any more.
The study of polymer solutions is often focused on the search of consti-
tutive laws for the viscous pressure tensor. One of the advantages of EIT is
to establish a connection between such constitutive equations and the non-
equilibrium equations of state derived directly by differentiating the expres-
sion of the extended entropy. These state equations are determinant in the
study of flowing polymer solutions, which is important in engineering, since
most of polymer processing take place under motion. The phase diagrams
established for equilibrium situations cannot be trusted in the presence of
202 7 Extended Irreversible Thermodynamics
Fig. 7.3 Phase diagram (temperature T vs. volume fraction φ) of a polymer solution
under shear flow. This is a binary solution of polymer polystyrene in dioctylphtalate
solvent for several values of P
v
12
(expressed in Nm
−2
). The dashed curve is the equi-
librium spinodal line (corresponding to a vanishing viscous pressure)
flows as the latter may enhance or reduce the solubility of the polymer and
the conditions under which phase separation occurs.
This explains why many efforts have been devoted to the study of flow-
induced changes in polymer solutions (Jou et al. 2000, 2001; Onuki 1997,
2002). Classical local equilibrium thermodynamics is clearly not a good
candidate because the equations of state should incorporate explicitly the in-
fluence of the flow. Moreover, the equilibrium thermodynamic stability con-
ditions cannot be extrapolated to non-equilibrium steady states, unless a
justification based on dynamic arguments is provided. According to EIT, the
chemical potential will explicitly depend on the thermodynamic fluxes, here
the viscous pressure. It follows that the physico-chemical properties related
to the chemical potential – as for instance, solubility, chemical reactions,
phase diagrams, and so on – will depend on the viscous pressure, and will
be different from those obtained in the framework of local equilibrium ther-
modynamics. This is indeed observed in the practice, see Fig. 7.3, where it is
shown that the critical temperature of phase change predicted by the equilib-
rium theory (281.4 K), as developed by Flory and Huggins, is shifted towards
higher values under the action of shear flow. The corrections are far from
being negligible when the shear is increased.
7.4 Microelectronic Devices
The classical thermodynamic theory of electric transport has been exam-
ined in Chap. 3. Here, we briefly discuss the EIT contribution to the study
of charge transport in submicronic electronic devices. Although the carrier
7.4 Microelectronic Devices 203
transport can always be described by means of the Boltzmann’s equation, to
solve it is a very difficult task and, furthermore, it contains more information
than needed in practical applications. The common attitude is to consider a
reduced number of variables (usually expressed as moments of the distribu-
tion function), which are directly related with density, charge flux, internal
energy, energy flux, and so on, and which are measurable and controllable
variables, instead of the full distribution function. This kind of approach is
referred to as a hydrodynamic model and EIT is very helpful in determining
which truncations among the hierarchy of evolution equations are compatible
with thermodynamics.
Before considering microelectronic systems, let us first study electric con-
duction in a rigid metallic sample. We assume that the electric current is
due to the motion of electrons with respect to the lattice. In CIT, the inde-
pendent variables are selected as being the specific internal u and the charge
per unit mass, z
e
; in EIT, the electric current i is selected as an additional
independent variable. For the system under study, the balance equations of
charge and internal energy may be written as
ρ ˙ z
e
= −∇ i , (7.67)
ρ ˙ u = −∇ q +i E, (7.68)
with E the electric field and i E the Joule heating term. Ignoring heat
transport (q = 0) for the moment, the generalized Gibbs’ equation takes the
form
ds = T
−1
du −T
−1
µ
e
dz
e
−αi di , (7.69)
with µ
e
being the chemical potential of electrons and α a phenomenologi-
cal coefficient independent of i . By following the same procedure as in the
previous sections, it is easily checked that the evolution equation for i is
τ
e
di
dt
= −(i −σ
e
E

), (7.70)
where E

= E−T∇(T
−1
µ
e
), τ
e
is the relaxation time, and σ
e
is the electrical
conductivity, provided that α in (7.69) is identified as α = τ
e
(ρσ
e
T)
−1
. The
generalized entropy is now given by
ρs = ρs
eq

τ
e

e
T
i i . (7.71)
Equation (7.70), i.e. a generalization of Ohm’s law i = σ
e
E, is often used
in plasma physics and in the analysis of high-frequency currents but without
any reference to its thermodynamic context.
A challenging application is the study of charge transport in submicronic
semiconductor devices for its consequences on the optimization of their func-
tioning and design. The evolution equations for the moments are directly
obtained from the Boltzmann’s equation. Depending on the choice of vari-
ables and the level at which the hierarchy is truncated, one obtains different
204 7 Extended Irreversible Thermodynamics
hydrodynamic models. A simple one is the so-called drift-diffusion model
(H¨ ansch 1991), where the independent variables are the number density of
electrons and holes, but not their energies. More sophisticated is the approach
of Baccarani–Wordeman, wherein the energy of electrons and holes is taken
as independent variables, but not the heat flux, assumed to be given by the
Fourier’s law. To optimize the description, a sound analysis of other possible
truncations is highly desirable. Application of EIT to submicronic devices
has been performed in recent works (Anile and Muscato 1995, Anile et al.
2003) wherein the energy flux rather than the electric flux is raised to the
level of independent variables. We will not enter furthermore into the details
of the development as they are essentially based on Boltzmann’s equation for
charged particles, which is outside the scope of this book. Let us simply add
that a way to check the quality of the truncation is to compare the predictions
of the hydrodynamic models with Monte Carlo simulations.
In particular, for a n
+
−n−n
+
silicon diode (Fig. 7.4) at room temperature,
the EIT model of Anile et al. (2003) provides results, which are in good
agreement with Monte Carlo simulations, as reflected by Fig. 7.5. Compared
to a Monte Carlo simulation, the advantage of a hydrodynamic model is its
much more economical cost with regard to the computing time consumption.
0.1mm 0.4 m m 0.1 m m
Fig. 7.4 A n
+
−n−n
+
silicon diode. The doping density in the region n
+
is higher
than in the region n
Fig. 7.5 Velocity profiles in the n
+
−n−n
+
silicon diode obtained, respectively, by
Monte Carlo simulations (dotted line) and the hydrodynamical model of Anile and
Pennisi (1992) based on EIT
7.5 Final Comments and Perspectives 205
7.5 Final Comments and Perspectives
To shed further light on the scope and perspectives of EIT, some general
comments are in form:
1. In EIT, the state variables are the classical hydrodynamic fields sup-
plemented by the fluxes provided by the balance laws, i.e. the fluxes
of mass, momentum, energy, electric charge, and so on. This attitude
is motivated by the fact that these dissipative fluxes are typically non-
equilibrium variables vanishing at equilibrium. The choice of fluxes is
natural as the only accessibility to a given system is through its bound-
aries. Moreover in processes characterized by high frequencies or systems
with large relaxation times (polymers, superfluids, etc.) or short-scale di-
mensions (nano- and microelectronic devices), the fluxes lose their status
of fast and negligible variables and find naturally their place among the
set of state variables. Other fields where the fluxes may play a leading
part are relativity, cosmology, traffic control (flux of cars), economy (flux
of money), and world wide web (flux of information). The choice of the
fluxes as variables finds its roots in the kinetic theory of gases. Indeed, it
amounts to selecting as variables the higher-order moments of the velocity
distribution function; in particular, taking the heat flux and the pressure
tensor as variables is suggested by Grad’s thirteen-moment theory (1958),
which therefore provides the natural basis for the development of EIT
(Lebon et al. 1992). The main consequence of elevating the fluxes to the
rank of variables is that the phenomenological relations of the classical
approach (CIT) are replaced by first-order time evolution equations of
Maxwell–Cattaneo type. In EIT, the field equations are hyperbolic; note,
however, that this property may not be satisfied in the whole space of
state variables, especially in the non-linear regime (M¨ uller and Ruggeri
1998; Jou et al. 2001). In CIT, the balance laws are parabolic of the dif-
fusion type with the consequence that signals move at infinite velocity.
EIT can be viewed as a generalization of CIT by including inertia in the
transport equations.
2. The space of the extra variables is not generally restricted to the above
ordinary dissipative fluxes. For instance, to cope with the complexity of
some fast non-equilibrium processes and/or non-local effects as in nano-
systems, it is necessary to introduce higher-order fluxes, such as the fluxes
of the fluxes, as done in Sect. 7.1.3. Moreover, it is conceivable that fluxes
may be split into several independent contributions, each with its own
evolution equation, as in non-ideal gases (Jou et al. 2001) and polymers
(see Sect. 7.3). In some problems, like those involving shock waves (Valenti
et al. 2002), it may be more convenient to use as variables combinations
of fluxes and transport coefficients.
3. Practically, it is not an easy task to evaluate the fluxes at each instant
of time and at every point in space. Nevertheless, for several problems of
practical interest, such as heat wave propagation, the fluxes are eliminated
206 7 Extended Irreversible Thermodynamics
from the final equations. Although the corresponding dispersion relations
may still contain the whole set of parameters appearing in the evolution
equations of the fluxes, like the relaxation times, the latter may however
be evaluated by measuring the wave speed, its attenuation, or shock prop-
erties. A direct measurement of the fluxes is therefore not an untwisted
condition to check the bases and performances of EIT.
4. There are several reasons that make preferable to select the fluxes rather
than the gradients of the classical variables (for instance, temperature
gradient or velocity gradient) as independent variables. (a) The fluxes are
associated with well-defined microscopic operators, and as such allow for
a more direct comparison with non-equilibrium statistical mechanics and
the kinetic theory. (b) The fluxes are generally characterized by short
relaxation times and therefore are more adequate than the gradients for
describing fast processes. Of course, for slow or steady phenomena, the use
of both sets of variables is equivalent because under these conditions the
former ones are directly related to the latter. (c) Expressing the entropy
in terms of the fluxes offers the opportunity to generalize the classical
theory of fluctuations and to evaluate the coefficients of the non-classical
part of the entropy as will be shown in Chap. 11. This would not be
possible by taking the gradients as variables. (d) Finally, the selection of
the gradients as extra variables leads to the presence of divergent terms
in the formulation of constitutive equations, a well-known result in the
kinetic theory.
5. EIT provides a strong connection between thermodynamics and dynam-
ics. In EIT, the fluxes are no longer considered as mere control parameters
but as independent variables. The fact that EIT makes a connection be-
tween dynamics and thermodynamics should be underlined. EIT enlarges
the range of applicability of non-equilibrium thermodynamics to a vast do-
main of phenomena where memory, non-local, and non-linear effects are
relevant. Many of them are finding increasing application in technology,
which, in turn, enlarges the experimental possibilities for the observation
of non-classical effects in a wider range of non-equilibrium situations.
6. It should also be underlined that EIT is closer to Onsager’s original con-
ceptualization than CIT. Indeed, according to Onsager, the fluxes are
defined as the time derivative of the state variables a
α
, and the forces are
given by the derivatives of the entropy with respect to the a
α
s
J
α
=
da
α
dt
, X
α
=
∂s
∂a
α
. (7.72)
Following Onsager, the time evolution equations of the a
α
s are obtained
by assuming linear relations between fluxes and forces
da
α
dt
=

β
L
αβ
X
β
. (7.73)
Now, the fluxes and forces of CIT are completely unrelated to Onsager’s
interpretation; clearly, the heat flux and the pressure tensor are not time
7.5 Final Comments and Perspectives 207
derivatives of state variables, similarly, the forces ∇T and V, widely used
in CIT, cannot be considered as derivatives of s with respect to the vari-
ables a
α
. Turning now back to EIT, one can define generalized fluxes J
α
and forces X
α
, respectively, by
J
q
=
dq
dt
, X
q
=
∂s
∂q
= αq, (7.74)
where q is the heat flux or any other flux variable and α is a phenomeno-
logical coefficient. Assuming now a linear flux–force relation J
q
= LX
q
,
with L = 1/ατ, one obtains an evolution equation for the state variables
q of the form
dq
dt
=
1
τ
(q
ss
−q), (7.75)
where q
ss
≡ −λ∇T is the classical Fourier steady state value of q. After
recognizing in (7.75) a Cattaneo-type relation, it is clear that the structure
of EIT is closer to Onsager’s point of view than that of CIT. Moreover,
by transposing Onsager’s arguments, it can be shown that the phenom-
enological coefficient L is symmetric (Lebon et al. 1992; Jou et al. 2001).
7. Extended irreversible thermodynamics is the first thermodynamic theory
which proposes an explicit expression for non-equilibrium entropy and
temperature. In most theories, this problem is even not evoked or the
temperature and entropy are selected as their equilibrium values, as for
instance in the kinetic theory of gases.
To summarize, the motivations behind the formulation of EIT were the
following:
• To go beyond the local equilibrium hypothesis
• To avoid the paradox of propagation of signals with an infinite velocity
• To generalize the Fourier, Fick, Stokes, and Newton laws by including:
– Memory effects (fast processes and polymers)
– Non-local effects (micro- and nano-devices)
– Non-linear effects (high powers)
The main innovations of the theory are:
• To raise the dissipative fluxes to the status of state variables
• To assign a central role to a generalized entropy, assumed to be a given
function of the whole set of variables, and whose rate of production is
always positive definite
Extended irreversible thermodynamics provides a decisive step towards a
general theory of non-equilibrium processes by proposing a unique formu-
lation of seemingly such different systems as dilute and real gases, liquids,
polymers, microelectronic devices, nano-systems, etc. EIT is particularly well
suited to describe processes characterized by situations where the product of
relaxation time and the rate of variation of the fluxes is important, or when
208 7 Extended Irreversible Thermodynamics
Table 7.2 Examples of application of EIT
High-frequency phenomena Short-wavelength phenomena
Ultrasounds in gases Light scattering in gases
Light scattering in gases Neutron scattering in liquids
Neutron scattering in liquids Heat transport in nano-devices
Second sound in solids Ballistic phonon propagation
Heating of solids by laser pulses Phonon hydrodynamics
Nuclear collisions Submicronic electronic devices
Reaction–diffusion waves in ecosystems Shock waves
Fast moving interfaces
Long relaxation times Long correlation lengths
Polyatomic molecules Rarefied gases
Suspensions, polymer solutions Transport in harmonic chains
Diffusion in polymers Cosmological decoupling eras
Propagation of fast crystallization fronts Transport near critical points
Superfluids, superconductors
the mean free path multiplied by the gradient of the fluxes is high; these situ-
ations may be found when either the relaxation times or the mean free paths
are long, or when the rates of change in time and space are high. Table 7.2
provides a list of situations where EIT has found specific applications.
It should nevertheless not be occulted that some problems remain still
open like:
1. Concerning the choice of state variables:
– Are the fluxes the best variables? Should it not be more judicious to
select a combination of fluxes or a mixing of fluxes and transport coef-
ficients?
– Where to stop when the flux of the flux and higher-order fluxes are taken
as variables? The answer depends on the timescale you are working
on. Shorter is the timescale, larger is the number of variables that are
needed.
– How far is far from equilibrium? In that respect, it should be convenient
to introduce small parameters related for instance to Deborah’s and
Knudsen’s numbers, allowing us to stop the expansions at a fixed degree
of accurateness.
2. What is the real status of entropy, temperature, and the second law far
from equilibrium?
3. Most of the applications concern fluid mechanics, therefore a description
of solid materials including polycrystals, plasticity, and viscoplasticity is
highly desirable.
4. The introduction of new variables increases the order of the basic differen-
tial field equations requiring the formulation of extra initial and boundary
conditions.
5. Turbulence remains a challenging problem.
7.6 Problems 209
It should also be fair to stress that, during the last decade, many efforts have
been spent to bring a partial answer to these acute questions.
7.6 Problems
7.1. Extended state space. Assume that the entropy s is a function of a vari-
able α and its time derivative η = dα/dt, and that α satisfies the differential
equation
M
d
2
α
dt
2
+

dt
= L
∂s
∂α
.
(a) Show that the positiveness of the entropy production demands that L > 0
but does not imply any restriction on the sign of M. Hint: Write dα/dt and
dη/dt in terms of ∂s/∂α and ∂s/∂η. (b) Assume that ∂s/∂η = aη, with
a being a constant. Show that the stability condition d
2
s < 0 implies that
M > 0 and a = −(M/L).
7.2. Phase velocity. Determine the expressions (7.8) of the phase velocity v
p
and the attenuation factor α. Hint: Substitute the solution T(x, t) =
T
0
exp[i(kx−ωt)] writtenintheformT(x, t) = T
0
exp[i Rek(x−v
p
t)] exp(−x/α)
in the hyperbolic equation (7.5) and split the result in real and imaginary parts.
7.3. Non-local transport. (a) Check that the entropy and entropy flux of the
non-local formalism including the flux of the heat flux presented in Sect. 7.1.3
are given by
ds =
1
T
du −
τ
1
v
λT
2
q dq −
τ
2
v
2λT
2
β

0
Q : d
0
Q−
τ
0
v
λT
2
β

QdQ
and
J
s
=
1
T
q +
1
λT
2
0
Q q +
1
λT
2
Qq.
(b) Prove that the entropy production corresponding to the Guyer–Krumhansl
equation (7.30) is

s
=
3
τ
R
ρc
v
c
2
0
q q +

N
5ρc
v
_
(∇q) : (∇q)
T
+ 2(∇ q)(∇ q)
¸
.
(c) Show that the stationary heat flux that satisfies the Guyer–Krumhansl
equation is the necessary condition for the entropy production to be a mini-
mum, under the constraint ∇ q = 0. In other terms, show that the Euler–
Lagrange equations corresponding to the variational equation
δ
_
(Tσ
s
−γ∇ q)dV = 0,
with respect to variations of q and γ are the steady state equations ∂u/∂t = 0
and ∂q/∂t = 0 provided one identifies the Lagrange multiplier γ with twice
the temperature (Lebon and Dauby 1990).
210 7 Extended Irreversible Thermodynamics
7.4. Non-local transport. Show that (7.26)–(7.28) can be obtained by writing
for q and Q general evolution equations of the form
˙ q = −∇ Π+σ
q
,
˙
Q = ∇ Ξ+ σ
Q
,
at the condition that the flux and source terms are given by the following
constitutive equations:
Π = A(T)I +QI +
0
Q, σ
q
= −q
and
Ξ = qI, σ
Q
= −Q,
wherein Ξ is a third-order tensor.
7.5. High-frequency wave speeds. (a) Obtain (7.32) from the energy balance
equation (7.2) and Guyer–Krumhansl relation (7.29), for longitudinal thermal
waves. (b) Verify that the high-frequency wave speed for longitudinal waves
is given by (7.33).
7.6. Continued-fraction expansions and generalized thermal conductivity. To
clarify the way to obtain the asymptotic expressions used in (7.36) and (7.40)
for the generalized thermal conductivity, (a) show that the continued fraction
R =
a
1 +
a
1+
a
1+...
tends, in the asymptotic limit of an infinite expansion, to
R

=
1
2
(

1 + 4a −1).
Hint: Note that, in this limit, R

must satisfy R

= a/(1 +R

). (b) From
this result, and assuming that all correlation lengths are equal, check that
(7.36) leads to (7.39) for steady state (namely ω = 0) and for k = 2π/L.
7.7. Phonon hydrodynamics. Poiseuille flow of phonons may be observed in
cylindrical heat conductors of radius R when the mean free paths
N
= c
0
τ
N
and
R
= c
0
τ
R
satisfy
N

R
¸ R
2
and
N
¸ R. In this case, (7.30) reduces
in a steady state to
∇T −
9
5
τ
N
ρc
v

2
q = 0.
(a) Show that the total flux along the cylinder will be given by
Q = −

24
ρc
v
τ
N
R
4
∇T.
7.6 Problems 211
(b) Compare the dependence of Q with respect to the radius R with the
corresponding expression obtained from Fourier’s law. Note that this
dependence may be useful to describe the decrease of the effective ther-
mal conductivity in very thin nanowires, in comparison with the usual
thermal conductivity of the corresponding bulk material.
7.8. Double time-lag behaviour. Instead of the Cattaneo’s equation (7.4),
some authors (Tzou 1997) use a generalized transport equation with two
relaxation times
τ
1
˙ q +q = −λ
_
∇T +τ
2

˙
T
_
.
(a) Introduce this equation into the energy balance equation (7.2) and obtain
the evolution equation for the temperature. (b) Discuss the limiting behaviour
of high-frequency thermal waves.
7.9. Two-temperature models. Many systems consist of several subsystems,
each with its own temperature, as for instance, the electrons and the lattice
in a metal. It has been shown that the evolution equations for the electron
and lattice temperatures T
e
and T
l
are given, respectively, by:
c
e
∂T
e
∂t
= ∇ (λ∇T
e
) −C(T
e
−T
l
),
c
l
∂T
l
∂t
= C(T
e
−T
l
).
The constant C describes the electron–phonon coupling, which accounts for
the energy transfer from the electrons to the lattice, and c
e
and c
l
are the
specific heats of the electrons and lattice per unit volume, respectively. (a)
When the solution of the first equation, namely T
e
= T
l
+ (c
l
/C)∂T
l
/∂t, is
introduced into the second one, prove that it leads to

2
T
l
+
c
l
C
∂∇
2
T
l
∂t
=
c
l
+c
e
λ
∂T
l
∂t
+
c
e
c
l
λC

2
T
l
∂t
2
.
(b) Show that this equation can also be obtained by eliminating q between the
energy balance equation and Guyer–Krumhansl’s equation, with the suitable
identifications of the parameters.
7.10. Limits of stability of non-equilibrium steady states. Let us first consider
heat conduction in a rigid solid for which s = s(u, q). (a) Write the second
differential δ
2
s of the generalized entropy (7.25) around a non-equilibrium
state with non-vanishing value of q. (b) Show that one of the conditions to
be satisfied in order that the matrix of the second differential be negative
definite is
α
c
v
T
2
+
_
α
2

2
α
∂u
2

_
∂α
∂u
_
2
_
q
2
≥ 0,
where α ≡ vτ/λT
2
. (b) If v, τ
1
, c
v
, and λ are constant, it is found that
∂α/∂u = −2α/(c
v
T) and ∂
2
α/∂u
2
= 6α(c
v
T)
−2
. Show that in this case the
former inequality is satisfied for values of q such that
212 7 Extended Irreversible Thermodynamics
q ≤
_
c
v
α
_
1/2
= ρc
v
T
_
λ
ρc
v
τ
1
_
1/2
,
where (λ/ρc
v
τ
1
)
1/2
= U is the second sound.
7.11. Development in gradients or in fluxes. Compare the behaviour of the
wave number-dependent viscosity η(k, ω) appearing in the two following
second-order expansions (a) the flux expansion
τ
2
∂P
v
∂t
+P
v
= −2ηV+∇
0
J
v
and
0
J
v
= −η

¸∇
0
P
v
),
with
0
J
v
the flux of the viscous pressure tensor, η

a phenomenological co-
efficient, and ¸. . .) the completely symmetrized part of the corresponding
third-order tensor and (b) the velocity gradient expansion
τ
2
∂P
v
∂t
+P
v
= −2ηV+
2

2
V.
Note that (b) yields an unstable behaviour for high values of k, since the
generalized viscosity becomes negative. (For a discussion of these instabili-
ties arising in kinetic theory approaches to generalized hydrodynamics, see
Gorban et al. 2004.)
7.12. Two-layer model and the telegrapher’s equation. The so-called two-layer
model consists of a system whose particles jump at random between two
states, 1 and 2, with associated velocities v
1
= v and v
2
= −v, respectively,
along the x-axis. Assume that the rate R of particle exchange between the
two states per unit time and length is proportional to the difference of the
probability densities P
1
and P
2
, i.e. R = r(P
1
−P
2
). Show that the evolution
equations for the total probability density P = P
1
+P
2
and for the probability
flux J = (P
1
−P
2
)v are, respectively,
∂P
∂t
+
∂J
∂x
= 0,
τ
∂J
∂t
+J = −D
∂P
∂x
,
with τ = 1/2r and D = v
2
τ (see van den Broeck 1990; Camacho and Zakari
1994).
7.13. Electrical system with resistance R and inductance L. The expression
relating the intensity I of the electrical current to the electromotive force ξ
is a relaxational equation given by
ξ = IR +L
dI
dt
,
7.6 Problems 213
with relaxation time τ
e
= L/R. The intensity I is related to the flux of electric
current i by I = iA, with A the cross section of the conductor. The magnetic
energy stored in the inductor is given by U
m
=
1
2
LI
2
. Consider the total
internal energy U
tot
= U + U
m
, with U the internal energy of the material.
Show that the Gibbs’ equation dS = T
−1
dU +T
−1
p dV may be rewritten as
dS = T
−1
dU +T
−1
p dV −
τ
e
V
σ
e
T
i di,
which is a relation reminiscent of the Gibbs’ equation proposed in EIT. Hint:
Recall that R = (σ
e
A)
−1
l, with l the length of the circuit.
7.14. Chemical potential. According to (7.66), the differential equation for
the Gibbs’ free energy at constant temperature and pressure reads as
dg =

k
µ
k
dc
k
+


P
v
: dP
v
,
and, after integration, g(c
i
, P
v
) = g
eq
(c
i
) +
1
4
vJP
v
: P
v
, where τ designates
the relaxation time of the viscous pressure tensor and J = η/τ, is the so-
called steady state compliance, a function which is often studied in polymer
solutions. (a) Obtain the modification to the chemical potential of the ith
component, defined by
µ
i
=
_
∂g
∂n
i
_
T,p,P
v
,
if the steady state compliance J is assumed to have the formJ = αM(ck
B
T)
−1
,
with M the molecular mass, c the number of macromolecules per unit vol-
ume, and α a constant (e.g. α = 0.400 for Rouse model of bead-and-spring
macromolecules, and α = 0.206 for the Zimm model, the latter including the
hydrodynamic interaction amongst the different beads of a macromolecule
(Doi and Edwards, 1986; Bird et al., vol. 2, 1987a, 1987b)). (b) Study the in-
fluence of the non-equilibrium contribution on the stability of the system; in
particular, determine whether the presence of a non-vanishing viscous pres-
sure will reinforce or not the stability with respect to that at equilibrium
(Jou et al. 2000).
Chapter 8
Theories with Internal Variables
The Influence of Internal Structure
on Dynamics
An alternative approach of non-equilibrium thermodynamics may be carried
out through the internal variables theory (IVT). This phenomenological de-
scription has been very successful in the study of a wide variety of processes in
rheology, deformable bodies, physico-chemistry and electromagnetism. IVT
may be considered as a generalization of the classical theory of irreversible
processes, just like extended irreversible thermodynamics. In the classical
theory, the state of the system is described by the local equilibrium vari-
ables, say energy, volume or deformation, but it is well known that many
non-equilibrium effects cannot be adequately described by this set of vari-
ables. To avoid, among others, the inconvenience of a heavy formalism in
terms of functionals, one has, in extended irreversible thermodynamics, sup-
plemented the local equilibrium variables by the corresponding fluxes. In the
IVT, one adopts a similar attitude by adding a certain number of internal
variables to the local equilibrium variables. Such variables are called internal
because they are connected either to internal motions or to local microstruc-
tures. They should not be confused with order parameters as introduced in
Chap. 6, which are describing some broken symmetry related to long-range
interactions as in phase transitions. We should also add that unlike the fluxes
in extended irreversible thermodynamics, the internal variables are arbitrary
extensive quantities, not identified from the outset. It is the freedom in the
choice of the internal variables that explains the wide domain of applicability
of the formalism.
What is really meant by internal variables? The usual interpretation is that
they are macroscopic representations of some microscopic internal structures;
they can be of geometrical or physico-chemical nature, but generally there is
a priori no reference to their origin. The hope is that at the end, some phys-
ical meaning will come out from the theory itself. Mandel (1978) wrote that
“a clever physicist will always manage to detect internal variables and mea-
sure them”. Internal variables are essentially introduced to compensate for
our lack of knowledge about the behaviour of the system. The alternative
name of hidden variables is also found in the literature, as they refer to the
215
216 8 Theories with Internal Variables
internal structure that is hidden from the eyes of an external observer. At the
difference of the external variables, like temperature or volume, the internal
parameters are not coupled to external forces which provide a means of con-
trol. As they cannot be adjusted to a prescribed value through surface and
body forces, internal variables are not controllable. Being not attached to an
external force, they cannot take part in a mechanical work and therefore they
do not appear explicitly in the balance equations of mass, momentum and
energy. Additional equations (taking generally the form of rate equations)
are needed to describe their evolution.
To summarize, internal variables are in principle measurable but not con-
trollable and this makes difficult their selection and definition. They are asso-
ciated to processes that are completely dissipated inside the system instead
of being developed against external surface or body forces.
Examples of internal variables are the extent of advancement in a chemical
reaction, the density of dislocations in plastic metals, mobile atoms in crys-
tals, the end-to-end distance between macromolecules in polymer solutions,
the shape and orientations of microstructures in macromolecules, the mag-
netization in ferromagnets, etc. In the following, we shall present the general
scheme of the thermodynamic theory of internal variables and afterwards we
shall discuss some applications both in fluid and solid mechanics.
Following our policy of seeking a presentation as simple as possible, we
shall restrict our analysis to rather simple examples. The reader interested
by more complete developments is advised to consult the comprehensive re-
view papers by Bampi and Morro (1984), Kestin (1990, 1992), Maugin and
Muschik (1994), Lhuillier et al. (2003), and the well-documented book by
Maugin (1999) which includes a large bibliography and a wide and system-
atic presentation.
8.1 General Scheme
8.1.1 Accompanying State Axiom
The hypothesis of local equilibrium that is the basis of classical irreversible
thermodynamics is replaced in the IVT by the accompanying state axiom.
The essence of this postulate is that, to each non-equilibrium state, corre-
sponds an accompanying equilibrium state, and to every irreversible process is
associated an accompanying “reversible” process. This notion can be traced
back to Meixner (1973a, 1973b), Muschik (1990) and mainly to Kestin (1990,
1992): it provides an interesting picture of IVT by associating non-equilibrium
states with equilibrium states by means of a projection. The accompanying
equilibrium state is a fictitious one; its state space 1 consists of the union of
the spaces of standard equilibrium variables ( and internal variables 1. The
8.1 General Scheme 217
ensemble 1 chosen for further analysis will consist of the following extensive
variables
1 := u, a, ξ; (8.1)
the space of classical variables includes the specific internal energy u and
the other local equilibrium variables, for instance the specific volume v in
fluid mechanics or the elastic strain tensor ε in solid mechanics, which will
be generically designed as a, and finally ξ ≡ ξ
1
, ξ
2
, . . . , ξ
n
stands for a finite
set of internal variables (either scalars, vectors, or tensors). Although the
number of variables needed to specify a non-equilibrium state n is generally
much larger than the set (8.1), we can always imagine an accompanying
equilibrium state e whose extensive variables u, a, ξ take the same values as
in a non-equilibrium state n, i.e. u(e) = u(n), a(e) = a(n), ξ(e) = ξ(n). A
way of obtaining these accompanying state variables from the actual physical
non-equilibrium state is illustrated by the Gedanken experience of Fig. 8.1.
Consider a volume element n suddenly surrounded by an adiabatic rigid
enclosure so that no heat flows and no external work can be performed;
as a consequence of the first law,
˙
Q =
˙
W = 0 implies that u(n) = u(e).
Simultaneously, we constrain the internal variables ξ. Since no work and no
heat are exchanged, the system relaxes towards an equilibrium-constrained
state e with the values of u, a, and ξ unmodified. Observe however that
other variables, like the temperature T or the entropy s, relax to values
which are different from their actual non-equilibrium values: T(e) ,= T(n) and
s(e) ,= s(n). The entropy s(e) is larger than the entropy in non-equilibrium
s(n) because the former is obtained from the latter by an adiabatic no-work
process. At this point it is interesting to emphasize the difference between the
axiom of accompanying state and the local equilibrium hypothesis. The gist
of this latter statement is to assume that the non-equilibrium entropy s(n)
and temperature T(n) are the same as in their accompanying equilibrium
state s(e) and T(e), respectively, while in IVT, the entropy s(e) is assumed
to characterize the equilibrium state.
Fig. 8.1 Producing an accompanying equilibrium e state from a non-equilibrium
state n
218 8 Theories with Internal Variables
Fig. 8.2 Accompanying reversible process
To clarify the notion of accompanying reversible process, consider an irre-
versible process between states n
1
and n
2
. The accompanying states e
1
and e
2
are obtained by projection on the state space (u, a, ξ) as shown in Fig. 8.2 and
the sequence of accompanying states between e
1
and e
2
constitutes the “re-
versible” accompanying process. By contrast with a true reversible process,
which takes place infinitely slowly, the former one occurs at finite velocity
and, as seen later, is characterized by a non-zero entropy production.
Now the question remains of how many internal variables are needed for a
valuable description. Of course it is desirable to have a number of variables as
small as possible. There exists a criterion based on the notion of characteristic
time that allows deciding whether an internal variable is relevant or not. Let
us denote by τ
ξ
=
˙
ξ/ξ the intrinsic time of the internal mechanism and by
τ
a
= ˙ a/a the characteristic macroscopic time associated with the external
variable a; as usual, a upper dot denotes time derivative. If the Deborah
number defined by De = τ
ξ

a
is such that De ¸1, then the corresponding
internal variable relaxes quickly towards its equilibrium value; this amounts
to say that the internal constraint is more or less removed, thus allowing
the system to reach a state of so-called unconstrained equilibrium (Kestin
1990, 1992). In contrast for De ¸1, the internal variable will remain frozen
during the timescale of the evolution of the variable a, we shall then speak
about a state of constrained equilibrium. Practically, in IVT only the internal
variables for which De ¸1 will be retained. It was also suggested by Kestin
(1990, 1992) to use the Deborah number as a measure of the “distance”
between a non-equilibrium state and an accompanying equilibrium state: the
smaller is De, the shorter is the distance between such states. Situations with
nonlocal effects are discussed in Box 8.1.
8.1 General Scheme 219
Box 8.1 Non-Local Effects. Histories of Variables
In presence of non-local effects, as in the cases of damages or non-
homogeneous electrical or magnetic fields, it may in principle be necessary
to introduce spatial gradients ∇ξ of the internal variables in the description.
The evolution of such internal variables is now determined by complete bal-
ance equations, involving both a rate and a divergence term. The difficulty
is that such balance laws require appropriate boundary conditions, which
cannot in principle be assigned, as the internal variables are not control-
lable. This problem has not received sufficient attention and only partial
answers can be found in the recent literature. The underlying idea is that,
although the internal variables are not controllable, the second law of ther-
modynamics forces them to assume given values on the boundaries (Woods
1975; Valanis 1996; Cimmelli 2002).
Because of the difficulty to identify the nature of the internal variables,
it may be tempting to formulate the problem without having recourse to
them. To obtain such a description, it suffices to eliminate the internal
variables ξ from (8.11) to (8.13). In the case that (8.13) can be solved,
we can substitute its solutions in (8.11) and (8.12), leading to a set of
constitutive relations, which turn out to be functionals Q and T of the
histories of u (or T), ∇T, and a (Kestin 1992):
q = Q
t

≥0
¦T(t −t

), ∇T(t −t

), a(t −t

)¦, (8.1.1)
F = T
t

≥0
¦T(t −t

), ∇T(t −t

), a(t −t

)¦. (8.1.2)
As shown in Chap. 9, it is customary in rational thermodynamics to claim
that the systems that it describe are materials with memory. The difficulty
is how to measure and to catalogue all these histories and how to solve the
subsequent integro-differential equations obtained after substitution of the
above functionals in the balance equations.
8.1.2 Entropy and Entropy Production
The axiom of accompanying equilibrium state leads us to accept the existence
of an accompanying entropy s that is a function of the whole set 1 of variables:
s = s(u, a, ξ). (8.3)
The corresponding Gibbs’ equation in rate form will assume the form
˙ s = T
−1
˙ u −T
−1
ρ
−1
F
e
˙ a +T
−1
A
˙
ξ, (8.4)
where T is the temperature, F
e
the mechanical force conjugate to the ob-
servable a, and A the affinity (also known as configurational or Eshelby
220 8 Theories with Internal Variables
force in some branches of solid mechanics) conjugate to ξ. They are defined,
respectively, by
1
T
=
∂s
∂u
,
F
e
ρT
= −
∂s
∂a
,
A
T
=
∂s
∂ξ
. (8.5)
The internal energy obeys the balance law
ρ ˙ u = −∇ q +F ˙ a (8.6)
with q the heat flux vector. Remark in passing that the forces F(n) required
to produce work on the volume element in the actual physical space are differ-
ent from the forces F
e
in (8.4) which represent these acting on the fictitious
reversible process. After elimination of ˙ u between the Gibbs’ equation (8.4)
and the energy balance equation (8.6), one obtains the expression for the
evolution of the entropy of the accompanying state during the process:
ρ˙ s +∇ (qT
−1
) = q ∇T
−1
+T
−1
(F −F
e
) ˙ a +ρT
−1
A
˙
ξ. (8.7)
Suppose that the entropy obeys a balance equation of the standard form
ρ˙ s +∇ (T
−1
q) = σ
s
≥ 0, (8.8)
wherein it is accepted that the rate of entropy production σ
s
is positive
definite and the entropy flux given by T
−1
q. By comparison with (8.7), it is
then found that
σ
s
= q ∇T
−1
+T
−1
(F −F
e
) ˙ a +ρT
−1
A
˙
ξ ≥ 0. (8.9)
With the exception of extended thermodynamics, it is currently admitted
that, in absence of diffusion of matter, the entropy flux is the ratio of the heat
flux and the temperature. The positiveness of the rate of entropy production is
a more disputed question and can be considered as a supplementary postulate
of the theory.
Instead of starting from (8.9), some authors prefer to work with the
Clausius–Duhem’s inequality
−ρ(
˙
f +s
˙
T) +F ˙ a +T
−1
q ∇T ≥ 0, (8.10)
which is directly obtained from (8.8) when use is made of the definition of
the free energy f = u − Ts and the energy balance equation (8.6). In the
future, we shall indifferently start from (8.9) or (8.10).
8.1.3 Rate Equations
It is noted that (8.9) has the form of the familiar sum of generalized forces
X and fluxes J, which, at the exception of the heat flux, are the rate of
change of the extensive variables. In classical irreversible thermodynamics
8.2 Applications 221
(see Chap. 2), the fluxes are linear expressions of the forces but this is not
mandatory in the present theory wherein the fluxes and forces are related
to each other by general functions containing the variables u, a, and ξ, i.e.
J = J(X; u, a, ξ). Such flux–force relations take generally the form of rate
equations, which conventionally will be written as
q = Q(∇T, ˙ a, A; u, a, ξ), (8.11)
F −F
e
= T(∇T, ˙ a, A; u, a, ξ), (8.12)
˙
ξ = Ξ(∇T, ˙ a, A; u, a, ξ), (8.13)
where Q, T, and Ξ denote functions of their arguments.
After this short digression, let us go back to the evolution equation (8.13).
In the simplest case, we may assume that the “flux”
˙
ξ is linear with the
“force” A
˙
ξ = lA(u, a, ξ), (8.13)
where l is a phenomenological coefficient. Such a simplified evolution equation
is justified when the thermal gradients play a negligible role and when F ≈ F
e
as often occurring in solid mechanics. The entropy production (8.9) takes now
the simple form

s
= ρA
˙
ξ, (8.14)
and we shall examine three particular cases:
1. A =
˙
ξ = 0: the system has reached a state of unconstrained equilibrium
characterized by A(u, a, ξ) = 0 or ξ(e) = ξ(u, a).
2. A ,= 0 with
˙
ξ = 0: this is a case of constrained equilibrium still defined
by a zero entropy production.
3. A ,= 0 and
˙
ξ ,= 0: the system relaxes towards the equilibrium surface
ξ(e).
As mentioned earlier, there exists a wealth of applications of the inter-
nal variable theory, mainly in the domain of material sciences. Here we will
present only three of them, and we refer the reader to specialized books and
articles for a more exhaustive list of examples.
8.2 Applications
8.2.1 Viscoelastic Solids
Viscoelastic solids are a privileged field of application of the theory of in-
ternal variables. Thermodynamics of viscoelasticity in solids was initiated
by Bridgman (1941) and Eckart (1948) and developed later on by Eringen
(1967), Kluitenberg (1962) and Kestin and Bataille (1980), among others.
Our purpose is to discuss qualitatively some essential features of rheological
222 8 Theories with Internal Variables
models like those of Maxwell and Kelvin–Voigt, which have played a central
role in the theory of viscoelasticity; excessive generality will be deliberately
avoided.
We restrict our attention to isothermal effects and make the assumption
that the solid is a one-dimensional homogeneous and isotropic rod subject
to infinitesimally small deformations. Under the action of a stress σ the solid
undergoes a strain ε. It is remembered from Sect. 8.1 that an intensive para-
meter such as σ, which is the stress actually applied in the non-equilibrium
state, is not the same as the stress σ
e
in the accompanying equilibrium state.
Moreover, it is assumed that the deformation of the body has, associated
with it, an internal work that is completely dissipated in the interior of the
system during the actual dissipative process. Its expression is given by /dξ
where / is the “affinity” associated to the internal variable ξ. According to
IVT, the space of basic variables is constituted by 1 ≡ u, σ
e
, ξ with u the
specific internal energy. The entropy associated with the accompanying state
satisfies the Gibbs’ equation
T ds = du −ρ
−1
σ
e
dε +/dξ, (8.15)
where ρ is the constant mass density. Since the temperature is a constant, it
may be preferable to work with the free energy f for which
df = ρ
−1
σ
e
dε −/dξ, (8.16)
with
ρ
−1
σ
e
= ∂f/∂ε, / = −∂f/∂ξ. (8.17)
The equation of state f = f(ε, ξ) can be expanded in a Taylor series around
a reference state ε = ξ = 0 and put in the form
f(ε, ξ) =
1
2

2
+Bεξ +
1
2

2
, (8.18)
where E, B, and C are constant coefficients with E > 0, C > 0, and B
2
≤ EC
to satisfy the stability property of the equilibrium state. Examples of linear
state equations (8.17) are
ρ
−1
σ
e
= Eε +Bξ, (8.19)
−/ = Bε +Cξ. (8.20)
Elimination of ˙ u from the energy balance equation
ρ ˙ u = σ : ˙ ε, (8.21)
and the Gibbs’ equation (8.15) written in rate form leads to the following
relation for the evolution of the entropy in the accompanying state:
ρT ˙ s = (σ −σ
e
) : ˙ ε +ρ
−1
A
˙
ξ. (8.22)
8.2 Applications 223
Since there is no heat flux, the entropy flux will be zero and therefore the
quantity ρT ˙ s can be identified with the rate of dissipated energy Tσ
s
, with
σ
s
the rate of entropy production. In very slow processes, it is justified to
assume (Kestin and Bataille 1980) that σ = σ
e
so that the dissipated energy
acquires the simple form
ρTσ
s
= A
˙
ξ ≥ 0. (8.23)
Of course, the simplest rate equation for the internal variable is to put
˙
ξ = lA(l > 0). (8.24)
Taking the time derivative of (8.19) and using (8.24), one obtains
ρ
−1
˙ σ = E ˙ ε +lB/, (8.25)
where use is made of the assumption σ
e
= σ. Substituting now / from (8.20)
and eliminating ξ from (8.19), leads to the following relaxation equation for
the stress tensor
τ
ε
˙ σ +σ = E

(ε +τ
σ
˙ ε), (8.26)
where τ
ε
= 1/lC, τ
σ
= E(lCE

)
−1
, E

= E −B
2
C
−1
. It is worth noticing
that relation (8.26) is the constitutive equation for a Poynting–Thomson
body. By letting τ
ε
and τ
σ
go to zero, one finds back the Hooke’s law of
elasticity, and E

is the classical Young modulus. Setting τ
ε
= 0, one recovers
the Kelvin–Voigt model
σ = E

ε +η ˙ ε, (8.27)
with η = E/lC, while by putting E

= 0 and E

τ
σ
= η in (8.26), one
obtains Maxwell’s model:
τ
ε
˙ σ +σ = η ˙ ε. (8.28)
It is therefore claimed that the thermodynamics of internal variables includes
the formulation of the standard models of rheology. The above considerations
have been generalized to finite strains by Sidoroff (1975, 1976), among others.
IVT has been used with success in several branches of solid mechanics. For
instance, an interesting approach of plasticity, in relation with the microscopic
theory of dislocations, was developed by Ponter et al. (1978). These authors
show that the area swept out by the dislocation lines is a meaningful internal
variable. More explicitly, the internal variable selected by Ponter et al. is
ξ =
ˆ
ANb

3
, (8.29)
where
ˆ
A is the area swept out by the dislocation line, N the number of non-
interacting so-called Franck–Read’s sources, b the Burgers vector associated
with dislocations slip, and
3
the corresponding volume. Another field of
application is the theory of fracture, where the internal variable can be chosen
as a scalar variable D (D for damage) such that 0 ≤ D ≤ 1, D = 0 meaning
absence of cracks and D = 1 corresponding to fracture.
224 8 Theories with Internal Variables
8.2.2 Polymeric Fluids
As a second illustration, we shall examine the behaviour of polymeric fluids,
which are one of the great diversity of fluids called complex fluids. Their
characteristic property is to possess a certain degree of internal structures
or microstructures. Therefore they behave very differently from water or air
when forces are exerted on them. The most rigorous method of describing
materials with internal microstructures is through the distribution function
ψ which depends on the position and velocity vectors of the ensemble of mass
points. However, in view of the large number of variables, this approach be-
comes quickly impracticable, and some mechanical approximations taking the
form of rigid rods, dumbbells or more complicated assemblages (see Fig. 8.3)
have been introduced.
In the case of dumbbells, the description involves a seven-dimensional func-
tion ψ(r, R, t) where r is the position vector and R the end-to-end macro-
molecular distance. But still more approximations are needed to simplify the
model. In that respect, it is frequent to replace the description based on the
distribution function by the second-order symmetric tensor
C =
_
RRψ dR, (8.30)
called “configuration” or “conformation” tensor. The advantage gained by
this approximation is to solve model equations in a large variety of flow sit-
uations. It is the conformation tensor that we shall identify as the relevant
internal variable for our forthcoming analysis. We consider a flowing incom-
pressible fluid composed of a carrier, the solvent of mass density ρ
f
, and long
macromolecular chains, the solute, of mass density ρ
pol
. Because of incom-
pressibility, the total mass ρ = ρ
f

pol
is a constant and ∇ v = 0, where v
designates the barycentric velocity defined by means of ρv = ρ
f
v
f

pol
v
pol
.
The selected basic variables are the mass concentration c
pol
= ρ
pol
/ρ of
the polymeric chain, the barycentric velocity v, the specific internal energy
u or the temperature T and the conformation tensor C. The corresponding
evolution equations are given by
Fig. 8.3 Molecular polymer models
8.2 Applications 225
ρ˙ c
pol
= −∇ J, (8.31)
ρ ˙ v = ∇ σ +ρF, (8.32)
ρ ˙ u = −∇ q + σ
v
: V, (8.33)
˙
C = ((C, V, c
pol
, u), (8.34)
wherein J denotes the flux of matter, F the external body force per unit
mass, V the symmetric velocity gradient tensor, ( a unknown function of the
variables C, c
pol
, u and eventually of V; finally, σ is the symmetric stress
tensor and σ
v
(= σ + pI) its viscous part with p the hydrostatic pressure.
The tensor σ
v
will be influenced by the presence of the internal variable
and therefore will depart from the Newton law by an additional contribution
so that in total
σ
v
= 2ηV+ σ
pol
(C, V, . . .), (8.35)
where η is the usual shear viscosity coefficient. The system of equations
(8.31)–(8.34) will be closed by providing constitutive relations for the set
¦J, q, σ
v
(or σ
pol
), C¦. These are obtained from the axiom of the local accom-
panying state consisting in formulating a generalized Gibbs’ equation and
calculating the corresponding rate of entropy production inside the material.
Let us write the Gibbs’ relation in terms of the free energy, i.e.
df = −s dT +µdc
pol
−A : dC, (8.36)
where µ = µ
pol
− µ
f
is the relative chemical potential, A is a symmetric
second-order tensor conjugate to the internal variable C describing the way
in which the free energy depends on the internal variable:
A = −
∂f
∂C
. (8.37)
There is no term of the form p d(ρ
−1
) in (8.36) because it is assumed that
the mass density is constant. By combining relation (8.36) with the balance
equations of mass concentration (8.31) and energy (8.33) and comparing with
the general evolution equation for the entropy
ρ˙ s = −∇ J
s

s
, (8.38)
one obtains the following results for the entropy flux and the entropy pro-
duction, respectively:
J
s
=
1
T
(q −µJ), (8.39)

s
= −J
s
∇T −J ∇µ + σ
v
: V+ρA :
˙
C ≥ 0. (8.40)
Expression (8.39) of the entropy flux is classical while the dissipated energy

s
contains a supplementary contribution compared to the similar relation
for a Newtonian fluid. For the time being, limiting ourselves to expressions of
226 8 Theories with Internal Variables
the fluxes J
s
, J, σ
v
, and
˙
C, which are linear and isotropic in the forces ∇T,
∇µ, V, and A, one is led to write
J
s
= −λ∇T −χ∇µ, (8.41)
J = −χ∇T −D∇µ, (8.42)
σ
v
= 2ηV+lA, (8.43)
˙
C = kA−lV = −k
∂f
∂C
−lV, (8.44)
wherein λ, χ, D, η, l, and k are phenomenological coefficients with λ > 0,
η > 0, D > 0, and k > 0. The same coefficient χ (respectively, l) appears
in (8.41) and (8.42) (respectively, in (8.43) and (8.44)) because of Onsager’s
reciprocal property, the change of sign in front of l in (8.43) and (8.44) is a
consequence of the property that the fluxes σ
v
and
˙
C are of opposite parity
with respect to time reversal. The terms lA in (8.43) and −lV in (8.44) do
not contribute to the entropy production and have therefore been labelled as
gyroscopic or non-dissipative. In most complex fluids, these non-dissipative
contributions may also contain non-linear terms (Lhuillier and Ouibrahim
1980) resulting from the coupling of A (or V) with C and (8.43) and (8.44)
will more generally be of the form:
σ
v
= 2ηV+lA−β(C A+A C), (8.45)
˙
C = k
∂f
∂C
−lV+β(C V+V C). (8.46)
The main results of the model are contained in (8.45) and (8.46) which show
that there is a strong interconnection between the way the molecular deforma-
tion C influences the stress tensor and the way the rate of fluid deformation
modifies the molecular deformation. A last remark is in form: if it is wished
to satisfy the axiom of material frame-indifference (see Chap. 9), the mater-
ial time derivative in (8.46), which is non-objective, must be replaced by an
objective one, as for instance the Jaumann time derivative
D
J
C =
˙
C+W C−C W, (8.47)
with W the skew-symmetric velocity gradient tensor. Moreover, there will
be no difficulty to generalize (8.45) and (8.46) with terms of third or higher
orders in C, V, and A, but we shall see that the present description offers
still enough possibilities (see Box 8.2).
Box 8.2 Simplified Models
Several simplifications of the model (8.45) and (8.46) are of interest:
(1) If in relation (8.46) we set β = 1 and = 0, one recovers a model
obtained in the kinetic theory. In this case the additional stress defined
by (8.35) is simply σ
pol
= 2A C, but the kinetic theory predicts
8.2 Applications 227
instead that σ
pol
≈ A from which we conclude that the results of the
kinetic theory are in contradiction with the IVT. The problem encoun-
tered in the kinetic theory finds probably its origin in the difficulties
arising in the evaluation of σ
pol
.
(2) Let us still assume that l = 0 but that the free energy f is quadratic
in C so that A = −∂f/∂C = −HC where H is some kind of Hookean
spring constant; (8.45) and (8.46) read then as
σ
v
= 2ηV+ 2βHC C, (8.2.1)
D
J
C = −
1
τ
C+β(C V+V C). (8.2.2)
This model was proved to be at least as valuable as the model of Oldroyd
(Bird et al. 1987) widely used in rheology; (8.2.1) exhibits clearly the
dependence of the non-Newtonian stress on the Hookean elasticity of
the polymer, whereas (8.2.2) is of the relaxation type with τ(= 1/kH), a
positive relaxation time. The physical interpretation of (8.2.2) is rather
evident: the presence of the Jaumann time derivative forces the mole-
cules to rotate with the surrounding fluid, the rate of fluid deformation
V deforms the molecules with an efficiency β and the molecules relax
towards an equilibrium configurational state with a relaxation time τ.
(3) Finally, by admitting that β = 0 and f quadratic in C, relation (8.46)
reduces to
D
J
C = −
1
τ
C−lV. (8.2.3)
Such an expression presents a striking resemblance with an equation
derived by Frankel and Acrivos (1970) to describe emulsions of quasi-
spherical drops with C a second-order tensor, expressing the deviation
of drops from sphericity.
8.2.3 Colloidal Suspensions
Colloidal suspensions are another example of complex fluids. A colloidal sus-
pension is a collection of particles suspended in a fluid, called the solvent or
the carrier. The driven particles must have a size larger than 10
−8
m for the
particles to be considered as a continuum, but smaller than 10
−4
m to avoid
sedimentation. They may be aerosols (fog, smog, smoke), emulsified drops,
ink, paint or asphalt particles. The suspending fluid must be sufficiently dense
in order that its mean free path be smaller than the size of the particles, thus
excluding rarefied gases. These hypotheses allow us to consider the system as
made by two interpenetrating continuous media. To simplify the issue, it is
generally admitted that the particles are rigid spheres of the same size mov-
ing in a Newtonian fluid so that the system may be considered as isotropic.
228 8 Theories with Internal Variables
In the foregoing analysis, we shall omit the role of velocity fluctuations whose
treatment can be found in specialized literature, e.g. Lhuillier (1995, 2001).
It is our aim to show that IVT provides a valuable tool for the study
of thermal diffusion in suspensions; the supramolecular size of the particles
being taken into account by the introduction of an internal variable. After
commenting in Sect. 8.2.3.1 the differences between particles suspensions and
molecular mixtures, we will present the evolution equations of the relevant
parameters and finally discuss the restrictions placed by the second law of
thermodynamics.
8.2.3.1 Suspensions vs. Molecular Mixtures
It is true that the suspensions of particles in a fluid behave in several aspects
as a binary molecular mixture of solute molecules in a solvent. However,
the particle size is larger than the molecular dimensions and this has two
important consequences. First, the volume of a particle is well defined whereas
the volume of a molecule is not. A description of suspensions will therefore
include not only the particle mass fraction c(= dM
p
/dM) but also its volume
fraction φ(= dV
p
/dV ) which is related to c by
φ = ρc/ρ
p
, (8.48)
where ρ is the total mass per unit volume of the suspension
ρ = φρ
p
+ (1 −φ)ρ
f
, (8.49)
the quantity ρ
p
(= dM
p
/dV
p
) is the mass per unit volume of the material from
which the particles are made and ρ
f
(= dM
f
/dV
f
) the mass per unit volume
of pure fluid; from now on, subscripts p and f will refer systematically to
“particle” and “fluid”, respectively. Moreover, to describe suspensions it is
necessary to introduce a volume-weighted velocity u defined by
u = φv
p
+ (1 −φ)v
f
, (8.50)
besides the classical barycentric mass-weighted velocity v,
v = cv
p
+ (1 −c)v
f
. (8.51)
The second difference between suspensions and molecular mixtures is the
large inertia of particles from which follows that the average particle velocity
v
p
can be very different from the fluid velocity v
f
. Hence the importance of
the particle diffusion flux
J = ρc(v
p
−v) = ρc(1 −c)(v
p
−v
f
), (8.52)
which is no longer given by the Fick’s law but will obey a full evolution
equation, as shown below. It is not difficult to convince oneself of the relations
8.2 Applications 229
u = v + (ρ
−1
p
−ρ
−1
f
)J, (8.53)
φ = c + (1 −c)(ρ
−1
p
−ρ
−1
f
), (8.54)
showing that u and φ are generally different from v and c respectively except
for the special cases that particles and fluid have similar mass densities.
8.2.3.2 State Variables and Evolution Equations
In presence of thermal gradients, the system will exhibit thermodiffusion.
Since we are in presence of two phases, a rather natural choice of variables
will be mass densities of phases, the average temperature and the velocities of
the particles and fluid phase. However, in analogy with molecular diffusion,
a more appropriate selection of variables is the following
ρ, c, T, v, ξ(= v
p
−v
f
), (8.55)
where the relative velocity will be considered as an internal variable. More
refined descriptions will require the introduction of more internal variables
as for instance a structural vector R related to the statistical distribution of
the particles (Lhuillier 2001) but, for the present purpose, it is sufficient to
restrict our analysis to the set (8.55). The evolution equations for the classical
variables ρ, c, T, and v are given by
˙ ρ = −ρ∇ v (total mass balance), (8.56)
ρ˙ c = −∇ J (mass concentration balance), (8.57)
ρ ˙ v = −∇p +∇ Π
v
+ρg(momentumbalance), (8.58)
ρ ˙ u = −∇ q −p∇ v +Π
v
: (∇v)
s
(internal energy balance), (8.59)
where p is the hydrostatic pressure, Π
v
is the symmetric viscous stress tensor,
(∇v)
s
the symmetric part of the tensor ∇v, and u the internal energy per unit
mass. It remains to determine the evolution equation of the internal variable
ξ, i.e. the relative velocity v
p
− v
f
. The latter is obtained from the two
momentum equations for the particulate and fluid phases given, respectively,
by Lhuillier (1995)
ρc
d
p
v
p
dt
= −φ(∇p −∇ σ
v
) +f +ρcg, (8.60)
ρ(1 −c)
d
f
v
f
dt
= −(1 −φ)(∇p −∇ σ
v
) −f −ρ(1 −c)g, (8.61)
where two new material time derivatives are introduced d
p
/dt = ∂/∂t +
v
p
∇ for the particles and a similar expression for the fluid; f is the mutual
interaction force between the particles and the fluid measured per unit volume
of the suspension. At this stage, f is an unknown quantity whose constitutive
230 8 Theories with Internal Variables
equation will be determined later on, σ
v
is the so-called suspension viscous
stress related to the stress Π
v
by Lhuillier (1995)
Π
v
= σ
v
−ξJ, (8.62)
besides the viscous term σ
v
, one observes the presence of a stress ξJ of kinetic
origin which is introduced to recover the global momentum equation (8.58).
Making use of the following identity
˙
ξ =
d
p
v
p
dt

d
f
v
f
dt
+∇
_
1
2
(2c −1)ξ
2
_
+ξ ∇v, (8.63)
one obtains the following evolution equation for the internal variable ξ:
˙
ξ = (ρ
−1
f
−ρ
−1
p
)(∇p−∇ σ
v
) +
f
ρc(1 −c)
+∇
_
1
2
(2c −1)ξ
2
_
−ξ ∇v. (8.64)
Set (8.56)–(8.59), (8.64) contains some indeterminate unknown quantities like
p, u, σ
v
, q, and f whose expressions will be determined from thermodynam-
ics and more particularly from the positive definite property of the entropy
production
σ
s
= ρ˙ s +∇ J
s
≥ 0. (8.65)
Of course at this stage of the discussion, the entropy s and the entropy flux
J
s
remain undetermined quantities to be expressed by means of constitutive
relations. Let us now examine the consequences issued from the positiveness
of σ
s
.
8.2.3.3 Restrictions Placed by the Second Law
of Thermodynamics
The total kinetic energy per unit mass can be written as
1
2
cv
2
p
+
1
2
(1 −c)v
2
f
=
1
2
v
2
+
1
2
c(1 −c)ξ
2
. (8.66)
To keep the usual expression
1
2
v
2
for the kinetic energy per unit mass, we
shall admit that the part of the kinetic energy involving the relative velocity
pertains to the internal energy so that
u(s, ρ, c, ξ) = u
0
(s, ρ, c) +
1
2
c(1 −c)ξ
2
, (8.67)
where u
0
is the local equilibrium energy depending exclusively on the set of
“equilibrium” variables. The corresponding Gibbs’ equation, written in rate
form, will therefore be given by
˙ u = T ˙ s +
p
ρ
2
˙ ρ +µ˙ c +
J
ρ

˙
ξ. (8.68)
8.2 Applications 231
Note that the chemical potential is now including a kinetic contribution and
is related to the local equilibrium chemical potential µ
0
(= ∂u
0
/∂c) by
µ = µ
0
+
1
2
(1 −2c)ξ
2
. (8.69)
After elimination of ˙ u, ˙ ρ, ˙ c, and
˙
ξ by means of the evolution equations (8.56)–
(8.59) and (8.64), we obtain the following entropy balance
T(ρ˙ s +∇ J
s
) ≡ Tσ
s
= ∇ [TJ
s
−q −µJ + (u −v) σ
v
] + σ
v
: (∇u)
sym
−J
s
∇T −J
_
∇µ
0
−(ρ
−1
p
−ρ
−1
f
)∇p +f [ρc(1 −c)]
−1
_
≥ 0. (8.70)
Positiveness of the dissipated energy Tσ
s
requires that the divergence term
in (8.70) vanishes, whence the following expression for the entropy flux:
J
s
=
1
T
(q −µJ) −
1
T
(u −v) σ
v
. (8.71)
The first two terms in (8.71) are classical but a new term depending on the
relative velocity and the mechanical stress tensor is appearing. The remaining
terms in (8.70) take the form of bilinear products of thermodynamic fluxes
and forces. The simplest way to guarantee the positive definite character of
the dissipated energy is to assume that these fluxes and forces are related by
means of linear relations, i.e.
σ
v
= η(∇u)
s
, (8.72)
J
s
= −
λ
T
∇T + ˜ sJ, (8.73)
f = −ρc(1 −c)
_
˜ s∇T −(ρ
−1
p
−ρ
−1
f
)∇p +∇µ
0
+D
−1
J
¸
, (8.74)
the phenomenological coefficients η, λ, ˜ s, D depend generally on ρ, c, and T,
the same ˜ s appears in both (8.73) and (8.74) to satisfy the Onsager reciprocal
relations. After substitution of the flux–force relations (8.72)–(8.74) in (8.70)
of the dissipated energy, one is led to

s
=
λ
T
(∇T)
2
+
1
D
J
2
+
1
η
σ
v
: σ
v
≥ 0, (8.75)
from which follows that λ > 0, η > 0, D > 0, there is no restriction on the
sign of ˜ s. The above results warrant further comments. It is important to note
that the stress σ
v
is related to the gradient of the volume-weighted velocity
u rather than to the gradient of the mass-weighted velocity v as in molecular
diffusion. This property has been corroborated by microscopic considerations
and is a well-known result in the theory of suspensions. Relation (8.73) can be
viewed as an expression of the Soret law stating that a temperature gradient
is capable of inducing a flux of matter.
The result (8.74) is important as it provides an explicit relation for the
inter-phase force f between the particles and the fluid, and plays, for sus-
pensions, the role of Fick’s law for binary mixtures. This interaction force
232 8 Theories with Internal Variables
will ultimately appear as a sum of elementary forces involving ∇c (through
(∂µ
0
/∂c)
p,T
) (concentration-diffusion force), ∇p (baro-diffusion force), ∇T
(thermodiffusion force) and the relative velocity v
p
− v
f
(through D
−1
J)
(kinematic-diffusion force).
Concerning the concentration-diffusion force, it always drives the particles
towards regions of lower particles concentration because (∂µ
0
/∂c)
p,T
> 0,
which is a consequence of thermodynamic stability. Experimental investiga-
tions confirm that the concentration-diffusion force is the most important,
that the thermodiffusion force is rather small, and that the baro-diffusion
force is negligible.
When the two following conditions are satisfied, ρ
p
= ρ
f
, and d
p
v
p
/dt =
d
f
v
f
/dt, it is found by subtracting (8.60) from (8.61) that the force f van-
ishes identically. If in addition the temperature and pressure are kept con-
stant, (8.74) boils down to Fick’s law J = −D∇µ
0
, where D is the positive
coefficient of diffusion.
Expression (8.74) of f is sometimes decomposed into a “non-dissipative”
force f

(the three first terms under brackets in (8.74)) and a dissipative
contribution, namely
f = f

−γ(v
p
−v), (8.76)
where use is made of the definition (8.52) of J and where γ, called the friction
coefficient, stands for γ = ρ
2
c
2
(1 − c)/D > 0. The term “non-dissipative” is
justified as it corresponds to situations for which γ = 0, i.e. D = ∞, which
is typical of absence of dissipation.
The expression of the heat flux vector q is directly derived by eliminating
J
s
between (8.71) and (8.73); making use of (8.72) and introducing a pseudo-
enthalpy function
˜
h = T ˜ s +µ, it is found that
q = −λ∇T +
˜
hJ +η(u −v) ∇u. (8.77)
For pure heat conduction, one recovers the classical Fourier law so that the
coefficient λ can be identified with the heat conductivity. For a molecular
mixture for which u = v, the above relation is equivalent to the law of
Dufour, expressing that heat can be generated by matter transport.
The above analysis shows that internal variables offer a valuable approach
of the theory of suspensions. It is worth noticing that the totality of results
obtained in this section was also derived in the framework of extended irre-
versible thermodynamics (Lebon et al. 2007).
8.3 Final Comments and Comparison
with Other Theories
Thermodynamics with IVT provides a rather simple and powerful tool for
describing structured materials as polymers, suspensions, viscoelastic bodies,
electromagnetic materials, etc. As indicated before, its domain of applicability
8.3 Final Comments and Comparison with Other Theories 233
is very wide, ranging from solid mechanics, hydrodynamics, rheology, electro-
magnetism to physiology or econometrics sciences. IVT requires only a slight
modification of the classical theory of irreversible processes by assuming that
the non-equilibrium state space is the union of two subsets. The first one is
essentially composed by the same variables as in classical irreversible ther-
modynamics while the second subset is formed by a more or less large set
of internal variables that have two main characteristics: first, they cannot be
controlled by an external observer and second, they can be unambiguously
measured. Furthermore, it is assumed that to any irreversible process, one
can associate a fictitious reversible process referred to as the accompanying
state. It was also proved that by eliminating one or several internal variables,
one is led to generalized constitutive relations taking the form of functional
of the histories of the state variables. In that respect, it can be said that the
IVT is equivalent to rational thermodynamics (see next chapter).
The main difficulty with IVT is the selection of the number and the iden-
tification of the nature of the internal variables. It is true that for some
systems, like polymers or suspensions, the physical meaning of these vari-
ables can be guessed from the onset, but this is generally not so. In most
cases, the physical nature of the internal variables is only unmasked at the
end of the procedure. In some problems, like in suspensions, the dependence of
the thermodynamic potentials on these extra variables is a little bit “forced”.
Referring for instance to (8.67) of the internal energy u, it is not fully justi-
fied that the dependence of u on the internal variable ξ is simply the sum of
the local internal energy and the diffusive kinetic energy. Another problem is
related to the time evolution of the internal variables. Except some particu-
lar cases, like diffusion of suspensions, there is no general technique allowing
us to derive these evolution equations, in contrast with extended irreversible
thermodynamics or GENERIC (see Chap. 10). Moreover, as these variables
are in principle not controllable through the boundaries, the evolution equa-
tions should not contain terms involving the gradients of the variables. This
is a limitation of the theory as it excludes in particular the treatment of non-
local effects. Some efforts have been recently registered to circumvent this
difficulty but the problem is not definitively solved. Unlike extended irre-
versible thermodynamics, where great efforts have been dedicated to a better
understanding of the notion of entropy and temperature outside equilibrium,
it seems that such questions are not of great concern in internal variables the-
ories. Here, the entropy that is used is the so-called accompanying entropy
and it is acknowledged that its rate of production is positive definite whatever
the number and nature of internal variables. The validity of such a hypothe-
sis if questionable and should be corroborated by microscopic theories as the
kinetic theory. The temperature is formally defined as the derivative of the
internal energy with respect to entropy but questions about the definition
of a positive absolute temperature and its measurability in systems far from
equilibrium are even not invoked. It is expected that the validity of the re-
sults of the IVT become more and more accurate as the number of internal
234 8 Theories with Internal Variables
variables is increased and would become rigorously valid when the number
of variables is infinite; however, from a practical point of view, this limit is
of course impossible to achieve.
8.4 Problems
8.1. Clausius–Duhem’s inequality. Show that the Clausius–Duhem’s inequal-
ity (8.10) is equivalent to the dissipation inequality (8.9).
8.2. Chemical reactions. Using the degree of advancement of a chemical reac-
tion ξ as an internal variable, formulate the problem of the chemical reaction
A + B = C + D in terms of the internal variable theory.
8.3. Particle suspensions. Why is the theory of molecular diffusion not ap-
plicable to the description of particle suspensions in fluids?
8.4. Viscoelastic bodies. Derive the constitutive relation (8.26) of a Poynting–
Thomson body by using Liu’s Lagrange multiplier technique developed in
Chap. 9.
8.5. Colloidal suspensions. Establish the evolution equation (8.64) of the in-
ternal variable ξ by using (8.60), (8.61), and (8.63).
8.6. Colloidal suspensions. Derive (8.74) of the interaction force f between
the particles and the fluid.
8.7. Colloidal suspensions. Eliminating the entropy flux between (8.71) and
(8.73) show that the heat flux in colloidal suspensions is given by
q = −λ∇T +
˜
hJ +η(u −v) ∇u.
In the particular problem of pure heat conduction, show that the above ex-
pression reduces to Fourier law, while for a mixture for which ρ
f
= ρ
p
(i.e.
u = v), it is equivalent to Dufour’s law.
8.8. Superfluids. Liquid He II is classically described by Landau’s two-fluid
model (see for instance Khalatnikov 1965). Accordingly, He II is viewed as a
binary mixture consisting of a normal fluid with a non-zero viscosity and a
superfluid with zero viscosity and zero entropy, the basic variables are ρ
n
, v
n
,
ρ
s
, v
s
, respectively, where ρ denotes the mass density and v the velocity field.
Show that an equivalent description may be achieved by selecting the relative
velocity ξ = (ρ
n
/ρ)(v
n
−v
s
) as an internal variable, with the corresponding
Gibbs’ equation given by
T d(ρs) = d(ρu) −g dρ −αξ d(ρξ),
wherein ρ = ρ
n
+ ρ
s
, α = ρ
s

n
while g = u − Ts + p(1/ρ) stands for the
specific Gibbs’ energy (see Lebon and Jou 1983; Mongiov`ı 1993, 2001).
8.4 Problems 235
8.9. Superfluids. Superfluid
4
He (see Lhuillier et al. 2003) is an ordered fluid
of mass per unit volume ρ and momentum per unit volume ρv; the latter is
understood as the sum of two contributions: one from the condensate driving
the total mass and moving with velocity v
s
, the other from elementary ex-
citations of momentum p and zero mass: ρv = ρv
s
+ p. The other original
feature of the superfluid is that it manifests itself by a curl-free velocity:
∇v
s
= 0.
Following the reasoning of Sect. 8.2, establish that the evolution equation for
p, considered as an internal variable, is given by
∂p/∂t +∇ [(v +c)p] + [∇(v +c)] p = −ρs∇T −ρ∇ψ
D
−∇ τ
D
,
where c is the variable conjugate to p/ρ, i.e. [c = −T∂s/∂(p/ρ)], ψ
D
the
dissipative part of Gibbs’ function g = ψ +ψ
D
, τ
D
the dissipative part of the
mechanical stress tensor.
8.10. Continuous variable. The internal variable ξ can also take the form of
a continuous variable with a Gibbs’ equation written as
T ds = du −p dv −
_
µ(ξ)dρ(ξ)dξ.
If the rate of change of ρ(x) is governed by a continuity equation ∂ρ/∂t =
−∂J(ξ)/∂ξ, which defines J(ξ) as a flux in the ξ-space, show that the corre-
sponding entropy production reads as

s
= −
_
J(ξ)∂µ(ξ)/∂ξ dξ ≥ 0,
suggesting integral phenomenological relations. However, if the internal vari-
able does not change abruptly, it is sufficient to require that only the inte-
grand of the above expression is positive so that, J(ξ) = −L∂ρ(ξ)/∂ξ.
8.11. Application to Brownian motion. In this problem, the internal variable
ξ will be identified as the x-component u of the Brownian particle velocity
(ξ = u), and the density ρ(ξ) represents the velocity distribution which, at
equilibrium is the Maxwellian one,
f
eq
= constant exp(−mu
2
/2k
B
T).
Assume that the potential µ(u) is of the form
µ(u) = (k
B
T/m) ln ρ(u) +A(u),
where µ
eq
= µ
0
is independent of u. Combining the two previous relations,
determine the explicit expression of A(u). Show that the phenomenological
relation can be cast in the form J = −L[f(u) − (k
B
T/m)∂f/∂u], where L
236 8 Theories with Internal Variables
is the friction coefficient of the Brownian particles. Combining this result
with the continuity relation, establish the Fokker–Planck equation for the
Brownian motion
∂f(u)
∂t
= L
_
∂f(u)
∂u
+
k
B
T
m

2
f(u)
∂u
2
_
.
8.12. Magnetizable bodies. In theories of magnetic solids under strain, it is
customary to select magnetization M = B − H (with B the magnetic in-
duction and H the magnetic field) as field variable and to split the mag-
netic variables into a reversible and an irreversible contribution, for instance,
M = M
r
+ M
i
, H
r
+ H
i
. However, to describe the complex relaxation
process, some authors (Maugin 1999, p 242) have introduced an extra inter-
nal variable M
int
. With this choice, the entropy production takes the form

s
= H
r

dM
r
dt
+H
i

dM
i
dt
+H
int

dM
int
dt
.
Show that H satisfies an evolution equation of the Cattaneo type
τ
dH
dt
+H=
τ
χ
m
dM
dt
,
where χ
m
denotes the magnetic susceptibility.
8.13. Vectorial internal variable and heat transport. Assume that the entropy
of a rigid heat conductor depends on the internal energy u and a vectorial in-
ternal variable j , i.e. s(u, j ). a) Obtain the constitutive equation for the time
derivative of j . b) From this equation, relate j to the heat flux q and express
the evolution equation for q, assuming, for simplicity, that all phenomenolog-
ical coefficients are constant; c) Compare this equation with the double-lag
equation presented in Problem 7.8. Which conditions are needed to reduce
it to the Maxwell-Cattaneo equation? Which form takes the entropy s(u, j )
when j is expressed in terms of q? Compare it with the extended entropy
(7.25).
Chapter 9
Rational Thermodynamics
A Mathematical Theory of Materials
with Memory
In Chaps. 7 and 8, it was assumed that the instantaneous local state of the
system out of equilibrium was characterized by the union of classical vari-
ables and a number of additional variables (fluxes in EIT, internal variables
in IVT). Only their instantaneous value at the present time was taken into
account and their evolution was described by a set of ordinary differential
equations. An alternative attitude, followed in the early developments of ra-
tional thermodynamics (RT), is to select a smaller number of variables than
necessary for an exhaustive description. The price to be paid is that the state
of the material body will be characterized not only by the instantaneous value
of the variables, but also by their values taken in the past, namely by their
history.
In RT, non-equilibrium thermodynamic concepts are included in a contin-
uum mechanics framework. The roots of RT are found in the developments
of the rational mechanics. Emphasis is put on axiomatic aspects with theo-
rems, axioms and lemmas dominating the account. Coleman (1964) published
the foundational paper and the name “rational thermodynamics” was coined
a few years later by Truesdell (1968). RT deals essentially with deformable
solids with memory, but it is also applicable to a wider class of systems in-
cluding fluids and chemical reactions. Its main objective is to put restrictions
on the form of the constitutive equations by application of formal statements
of thermodynamics. A typical feature of RT is that its founders consider it
as an autonomous branch from which it follows that a justification of the
foundations and results must ultimately come from the theory itself. A vast
amount of literature has grown up about this theory which is appreciated
by the community of pure and applied mathematicians attracted by its ax-
iomatic vision of continuum mechanics.
In the present chapter we present a simplified “idealistic” but neverthe-
less critical version of RT laying aside, for clarity, the heavy mathematical
structure embedding most of the published works on the subject.
237
238 9 Rational Thermodynamics
9.1 General Structure
The basic tenet of rational thermodynamics is to borrow those notions and
definitions introduced in classical thermodynamics to describe equilibrium
situations and to admit a priori that they remain applicable even very far
from equilibrium. In that respect, temperature and entropy are considered as
primitive concepts which are a priori assigned to any state. Quoting Truesdell
(1984), it is sufficient to know that “temperature is a measure of how hot a
body is, while entropy, sometimes called the caloric, represents how much
heat has gone into a body from a body at a given temperature”.
Similarly, the second law of thermodynamics written in the form ∆S ≥
_
¯ dQ/T and usually termed the Clausius–Duhem’s inequality is always sup-
posed to hold. It is utilized as a constraint restricting the range of acceptable
constitutive relations. The consequence of the introduction of the history is
that Gibbs’ equation is no longer assumed to be valid at the outset as in the
classical theory of irreversible thermodynamics. Since the Gibbs equation is
abandoned, the distinction between state equations and phenomenological re-
lations disappears, everything will be collected under the encompassing word
of constitutive equations. Of course, the latter cannot take any arbitrary
forms as they have to satisfy a series of axioms, most of them being elevated
to the status of principles in the RT literature.
9.2 The Axioms of Rational Thermodynamics
To each material is associated a set of constitutive equations specifying partic-
ular properties of the system under study. In RT, these constitutive relations
take generally the form of functionals of the histories of the independent
variables and are kept distinct from the balance equations. In the present
chapter, the latter turn out to be
˙ ρ = −ρ∇ v (mass balance), (9.1a)
ρ ˙ v = ∇ σ +ρF (momentum balance), (9.1b)
ρ ˙ u = −∇ q +σ : ∇v +ρr (internal energy balance). (9.1c)
As in the previous chapters, a superimposed dot stands for the material
time derivative, ρ is the mass density; u, the specific internal energy; v, the
velocity; and q is the heat flux vector; in rational thermodynamics, it is
preferred to work with the symmetric Cauchy stress tensor σ instead of the
symmetric pressure tensor P(= −σ). It is important to observe the presence
of the specific body force F in the momentum equation and the term r
in the energy balance, which represents the energy supply due to external
sources, for instance the energy lost or absorbed by radiation per unit time
9.2 The Axioms of Rational Thermodynamics 239
and unit mass. It must be realized that the body force and the source term
are essentially introduced for the self-consistency of the formalism. Contrary
to the classical approach, F and r are not quantities which are assigned a
priori, but instead the balance laws will be used to “define” them, quoting
the rationalists. In other terms, the balance equations of momentum and
energy are always ensured as we have two free parameters at our disposal.
The quantities F and r do not modify the behaviour of the body and do
not impose constraints on the set of variables, but rather, it is the behaviour
of the material, which determines them. This is a perplexing attitude, as F
and r, although supplied, will always modify the values of the constitutive
response of the system.
The principal aim of RT is to derive the constitutive equations character-
izing a given material. Of course, these relations cannot take arbitrary forms,
as they are submitted to a series of axioms, which place restrictions on them.
Let us briefly present and discuss some of these most relevant axioms.
9.2.1 Axiom of Admissibility and Clausius–Duhem’s
Inequality
By “thermodynamically admissible” is understood a process whose constitu-
tive equations obey the Clausius–Duhem’s inequality and are consistent with
the balance equations. As will see later, the Clausius–Duhem’s inequality
plays a crucial role in RT. The starting relation is the celebrated Clausius–
Planck’s inequality, found in any textbook of equilibrium thermodynamics,
and stating that between two equilibrium sates A and B, one has
∆S ≥
_
B
A
¯ dQ/T. (9.2)
Since the total quantity of heat ¯ dQ results from the exchange with the exte-
rior through the boundaries and the presence of internal sources, the above
relation may be written as
d
dt
_
V
ρs dV ≥ −
_
Σ
1
T
q n dΣ +
_
V
ρ
r
T
dV , (9.3)
where s is the specific entropy, V is the total volume, and n is the outwards
unit normal to the bounding surface Σ. In local form, (9.3) writes as
ρ˙ s +∇
q
T

ρr
T
≥ 0. (9.4)
It is worth to note that the particular form (9.4) of the entropy inequality
is restricted to the class of materials for which the entropy supply is given
by ρr/T and the entropy flux by q/T. For a more general expression of
240 9 Rational Thermodynamics
the entropy flux, see extended irreversible thermodynamics (Chap. 7). After
elimination of r between the energy balance equation (9.1c) and inequality
(9.4) and introduction of Helmholtz’s free energy f(= u−Ts), one comes out
with
−ρ(
˙
f +s
˙
T) +σ : ∇v −q
∇T
T
≥ 0, (9.5)
which is referred to as the Clausius–Duhem’s or the fundamental inequality. It
is easily checked (see Problem 9.1) that the left-hand side of (9.5) represents
the rate of dissipated energy Tσ
s
per unit volume when the entropy flux is
given by q/T.
9.2.2 Axiom of Memory
If it is admitted that the present is influenced not only by the present state
but also by the past history, the constitutive relations will depend on the
whole history of the independent variables. If ϕ(t) designates an arbitrary
function of time, say the temperature or the strain tensor, its history up to
the time t is defined by ϕ
t
= ϕ(t −t

) with 0 ≤ t

< ∞.
The axiom of memory asserts that the behaviour of the system is com-
pletely determined by the history of the set of selected independent variables.
This means that the free energy, the entropy, the heat flux and the stress
tensor, for instance, will be expressed as functionals of the history of the
independent variables. Considering the problem of heat conduction in a
rigid isotropic material, an example of constitutive equation with memory
is Fourier’s generalized law
q(t) =
_
t
−∞
λ(t −t

)∇T(t

)dt

, (9.6)
where λ(t − t

) is the memory kernel. When this expression is substi-
tuted in the energy balance, one obtains an integro-differential equation
for the temperature field, after use is made of ˙ u = c
˙
T with c the specific
heat capacity. If the memory kernel takes the form of an exponential like
(−λ/τ) exp[−(t − t

)/τ], it is left as an exercise (see Problem 9.2) to show
that the time derivative of (9.6) is given by
τ ˙ q = −q −λ∇T, (9.7)
which is the same Cattaneo equation as in EIT. It is important to realize that
this result has been obtained by considering only the temperature as single
state variable. This is a characteristic of RT where the state space is generally
restricted to the classical variables, i.e. mass, velocity (or deformation), and
temperature, while the fluxes are expressed in terms of integral constitutive
9.2 The Axioms of Rational Thermodynamics 241
equations containing the whole history of the independent variables. Instead
of assuming that q depends on the whole history of temperature field, in
practical applications it is assumed that q is a function of ∇T and its higher-
order time derivatives. If the memory is very short in time, one may restrict
this sequence to a limited number of terms. But even in this case, RT offers
an interesting formalism which departs radically from that of classical
irreversible thermodynamics.
In most situations, the notion of fading memory is also introduced. Accord-
ingly, the distant history has little influence on the present state; although
history is often described by an exponentially decreasing function of time,
it could take more general forms as a sum of exponentials or of Gaussian
memories. However, to avoid heavy mathematical developments, we shall
suppose from now on that the materials forget their past experience quasi-
instantaneously so that memory effects can be neglected.
In some versions of RT, the description in terms of histories is substituted
by the state-process formalism (Noll 1974; Coleman and Owen 1974). Fol-
lowing these authors, to each thermodynamical system is associated a pair
formed by an instantaneous state and a process describing the temporal evo-
lution of the state space. The methods, tools and prospects of this theory
present similar features with these described in this chapter; an exhaustive
analysis of Coleman and Owen’s approach can also be found in Silhavy (1997).
9.2.3 Axiom of Equipresence
This axiom states that if a variable is present in one constitutive relation,
then there is no reason why it should not be present in all the other con-
stitutive equations, until it is proved otherwise. The condition for the pres-
ence or absence of an independent variable is essentially determined by the
Clausius–Duhem’s inequality. It should be realized that there is no physical
justification to such an axiom, which is merely a mathematical convenience
in the determination of constitutive relations.
9.2.4 Axiom of Local Action
It is admitted that a material particle is only influenced by its immediate
neighbourhood and that it is insensitive to what happens at distant points.
Practically, it means that second and higher-order space derivatives are ex-
cluded from the constitutive relations. Higher spatial gradients have however
been included in some developments of the theory on non-local actions.
242 9 Rational Thermodynamics
9.2.5 Axiom of Material Frame-Indifference
Generally speaking, this axiom claims that the response of a system must
be independent of the motion of the observer. As the most general motion
of an observer, identified as a rigid coordinate system, is constituted by a
translation and a rotation, the axiom implies that the constitutive equations
must be invariant under the Euclidean transformation
x

(t) = Q(t) x(t) +c(t). (9.8)
The quantity Q(t) is an arbitrary, real, proper orthogonal, time-valued tensor
satisfying
Q Q
T
= Q
T
Q = I, det Q = 1, (9.9)
c(t) is an arbitrary time-dependent vector; x(t), the position vector of a
material point at the present time and x

(t) is the position occupied after
having undergone a rotation (first term in the right-hand side of (9.8)) and
a translation (second term in the right-hand side). In the particular case
Q = I and c(t) = v
0
t with v
0
a constant velocity, (9.8) reduces to a Galilean
transformation.
When the Euclidean group (9.8) acts on a tensor of rank n(n = 0, 1, 2),
the latter is said to be objective if it transforms according to
a

= a (objective scalar), (9.10a)
a

= Q a (objective vector), (9.10b)
A

= Q A Q
T
(objective tensor). (9.10c)
We directly observe that the velocity v

= dx

/dt is not an objective vector
as it transforms as
v

= Q v +
˙
Q x + ˙ c,
which is not of the form (9.10b) because of the presence of the second and
third terms in the right-hand side of the above relation; the same is true for
the acceleration, which is not objective. It is easily checked (see Problems 9.3
and 9.4) that the velocity gradient, the angular velocity tensor and the ma-
terial time derivative of objective vectors or tensors are non-objective; in
contrast, quantities like the temperature gradient, the mass density gradient
or the symmetric velocity gradient tensor are objective.
We are now in position to propose a more precise formulation of the princi-
ple of material frame-indifference which rests on the following requirements.
First, the primitive variables as temperature, energy, entropy, free energy
and energy supply are by essence objective scalars, the body force and the
heat flux are objective vectors while the stress tensor is an objective ten-
sor. Second, the constitutive relations are objective, i.e. form invariant with
respect to the Euclidean transformation (9.8). Third, the constitutive equa-
tions, which reflect the material properties of a body, cannot depend on the
9.3 Application to Thermoelastic Materials 243
angular velocity of the reference frame. To give an example, the Newton’s
equation of rational mechanics, when formulated in a non-inertial rotating
frame, is Euclidean invariant but it depends explicitly on the angular ve-
locity of the frame and therefore, it does not fulfil the axiom of material
frame-indifference.
9.3 Application to Thermoelastic Materials
Consider a deformable, anisotropic elastic solid. Under the action of exter-
nal forces and heating, the material changes its configuration from a non-
deformed one with mass density ρ
0
to a deformed state with mass density
ρ. The position of the material points is denoted by X, in the non-deformed
configuration and by x = χ(X, t), in the deformed one with u = x −X the
displacement vector.
Loyal to our principle of simplicity, we shall restrict the analysis to linear
thermoelasticity, i.e. small deformations and small temperature increments
T −T
0
with respect to a reference temperature T
0
. A more general description
implying large deformations is treated in Box 9.1. Within the limit of small
deformations, the density ρ remains constant and the balance equations for
the displacement vector u and the temperature field read as (Eringen 1967;
Truesdell and Toupin 1960)
ρ¨ u = ∇ σ +ρF, (9.11)
ρ ˙ u = −∇ q +σ : ˙ ε +ρr, (9.12)
where ε =
1
2
[∇u + (∇u)
T
] stands for the symmetric strain tensor. These
relations contain unknown quantities as u (or equivalently f), σ and q
which must be specified by constitutive equations, compatible with Clausius–
Duhem’s inequality. In (9.12), the scalar u (internal energy) should not be
confused with the vector u (displacement vector) appearing in (9.11).
A thermoelastic material is defined by the following constitutive equations
f = f(T, ∇T, ε), (9.13)
s = s(T, ∇T, ε), (9.14)
σ = σ(T, ∇T, ε), (9.15)
q = q(T, ∇T, ε), (9.16)
where use has been made of the axiom of equipresence, s has been included
among the constitutive relations as it figures explicitly in Clausius–Duhem’s
relation. By no means are the above constitutive equations the most general
that one could propose but they appear as particularly useful to describe a
large class of deformable elastic solids. As a consequence of (9.13) and using
the chain differentiation rule, one can write the time derivative of f as
244 9 Rational Thermodynamics
˙
f =
∂f
∂T
˙
T +
∂f
∂(∇T)

.
∇T +
∂f
∂ε
: ˙ ε. (9.17)
It is left as an exercise (see Problem 9.5) to prove that the Clausius–Duhem’s
inequality (9.5) will take the form
−ρ(
˙
f +s
˙
T) +σ : ˙ ε −
q ∇T
T
≥ 0, (9.18)
and, after substitution of (9.17),
−ρ
_
s +
∂f
∂T
_
˙
T −
_
ρ
∂f
∂ε
−σ
_
: ˙ ε −ρ
∂f
∂∇T

.
∇T −q
∇T
T
≥ 0, (9.19)
which is obviously linear in
˙
T, ˙ ε, and
·
∇T. Moreover, since there exist body
forces and energy supplies that ensure that the balance equations of momen-
tum and energy are identically satisfied, these laws do not impose constraints
on
˙
T, ˙ ε, and
·
∇T, which can therefore take arbitrary prescribed values. In or-
der that the entropy inequality (9.19) holds identically, it is then necessary
and sufficient that the coefficient of each time derivative vanishes. As a con-
sequence, it follows that
s = −∂f/∂T, (9.20a)
σ = ρ∂f/∂ε, (9.20b)
∂f/∂(∇T) = 0. (9.20c)
It is concluded from (9.20c) that the free energy f does not depend on the
temperature gradient and on account of (9.20a) and (9.20b), the same obser-
vation holds for the entropy s and the stress tensor σ so that (9.13)–(9.15)
will take the form
f = f(T, ε), s = s(T, ε), σ = σ(T, ε). (9.21)
From (9.20a), (9.20b), and (9.20c), we can write the differential expression
df = −sdT +ρ
−1
σ : dε (9.22)
or, equivalently,
T ds = du −ρ
−1
σ : dε, (9.23)
which is the Gibbs equation for thermoelastic bodies. Note that this relation
has not been assumed as a starting point but has been derived within the
formalism.
Expanding f up to the second order in σ and T − T
0
, one obtains, in
Cartesian coordinates, and using Einstein’s summation convention,
f =
1

C
ijkl
ε
ij
ε
kl

1
2
c
ε
T
0
(T −T
0
)
2

1
ρ
β
ij
ε
ij
(T −T
0
), (9.24)
9.3 Application to Thermoelastic Materials 245
where C
ijkl
is the fourth-order tensor of elastic moduli; c
ε
, the heat capacity
and β
ij
is the second-order tensor of thermal moduli. In virtue of (9.20a)
and (9.20b), the corresponding linear constitutive equations of s and σ
ij
are
given by
s =
c
ε
T
0
(T −T
0
) +
1
ρ
β
ij
ε
ij
, (9.25)
σ
ij
= C
ijkl
ε
kl
−β
ij
(T −T
0
). (9.26)
The result (9.26) is the well-known Neumann–Duhamel’s relation of ther-
moelasticity, which simplifies to Hooke’s law when the temperature is uniform
(T = T
0
).
Going back to Clausius–Duhem’s expression (9.19), the latter reduces to
the remaining inequality
q(T, ∇T, ε) ∇T ≤ 0, (9.27)
which reflects the property that heat flows spontaneously from high to low
temperatures. Defining the heat conductivity tensor by
K(T, ε) = −
_
∂q(T, ∇T, ε)
∂(∇T)
_
∇T=0
(9.28)
and expanding q around ∇T = 0 with T and ε fixed, one obtains in the
neighbourhood of ∇T = 0, i.e. by omitting non-linear terms,
q(T, ∇T, ε) = q(T, 0, ε) −K(T, ε) ∇T. (9.29)
Substitution of (9.29) in (9.27) yields,
q(T, 0, ε) ∇T −∇T K(T, ε) ∇T ≤ 0. (9.30)
Since this relation must be satisfied for all ∇T, it is required that
q(T, 0, ε) = 0, ∇T K(T, ε) ∇T ≥ 0. (9.31)
From this, it is immediately concluded that in an admissible thermoelastic
process, the heat flux is zero when the temperature gradient vanishes and the
heat conductivity tensor, which is independent of ∇T, is positive definite. In
fact, only the symmetric part K
sym
= K + K
T
of K enters the statement
(9.31b) to be positive definite, this is so because for the skew part K
skew
=
K−K
T
, one has ∇T K
skew
∇T = 0. It follows from the above considerations
that expansion (9.29) becomes
q(T, ∇T, ε) = −K(T, ε) ∇T. (9.32)
This is a generalization of Fourier’s equation in which the thermal conduc-
tivity is a function not only of the temperature but also of the strain tensor.
246 9 Rational Thermodynamics
We observe also from (9.32) that in a thermoelastic solid, it is not possible
to produce a heat flux by a deformation only, at uniform temperature. This
means that the presence of a piezoelectric effect is excluded in thermoelastic
bodies.
The generalization of the above results to large deformations is straight-
forward and is presented in the Box 9.1.
Box 9.1 Finite Deformations in Thermoelastic Materials
It is usual in solid mechanics to introduce the following quantities (e.g.
Eringen 1962):
F = ∇χ (deformation tensor, F
ij
= ∂x
i
/∂X
j
in Cartesian coordi-
nates, with x
i
and X
j
the position in the deformed and non-
deformed reference configuration, respectively),
E =
1
2
(F
T
F −I) (Green symmetric strain tensor),
L =
˙
F F
−1
= ∇v (strain rate tensor, (∇v)
ij
= ∂v
j
/∂x
i
in
Cartesian coordinates),
V = symL (symmetric part of L),
T = (ρ
0
/ρ)σ F
−T
(first Piola–Kirchhoff stress tensor),
S = F
−1
T (symmetric second Piola–Kirchhoff stress tensor),
Q = (ρ
0
/ρ)F
−1
q (heat flux measured in the reference non-
deformed configuration).
The balance laws of mass, momentum and energy take the form
ρ
0
/ρ = det F,
ρ
0
˙ v = ∇ T+ρ
0
F,
ρ
0
˙ u = −∇ q +S :
˙
E+ρr.
The results derived for small deformations are still valid at the condition
to replace everywhere ε by E, σ by S, q by Q, and ∇T by G = F
T
∇T,
the temperature gradient with respect to the reference configuration. For
example, the Clausius–Duhem’s inequality reads as
−ρ
0
_
∂f
∂T
+s
_
˙
T −
_
ρ
0
∂f
∂E
−S
_

˙
E−ρ
0
∂f
∂G

˙
G −Q
G
T
≥ 0,
leading to the restrictions,
∂f/∂T = −s, ∂f/∂E = S/ρ
0
, ∂f/∂G = 0,
with a generalized Fourier law given by
Q(T, G, E) = −K(T, E) G.
9.4 Viscous Heat Conducting Fluids 247
Although most of the applications of RT have been devoted to solid mechan-
ics, there is no problem to adapt the above considerations to hydrodynamics,
as shown in Sect. 9.4.
9.4 Viscous Heat Conducting Fluids
To avoid lengthy mathematical developments with constitutive equations in
functional form, we shall consider a particular class of fluids characterized by
isotropy, absence of memory and described by the following set of constitutive
equations:
f = f(v, T, V, ∇T), (9.33)
s = s(v, T, V, ∇T), (9.34)
σ = σ(v, T, V, ∇T), (9.35)
q = q(v, T, V, ∇T), (9.36)
wherein v = ρ
−1
is the specific volume, V =
1
2
[∇v + (∇v)
T
] the symmetric
velocity gradient tensor. Note that neither the velocity field v nor the ve-
locity gradient ∇v have been included in the set of independent variables,
because they are not objective quantities and therefore do not satisfy the
axiom of frame-indifference. Instead, the tensor V has been selected as it
meets the property to be objective. Observe in passing that the above re-
sponse functions are in full agreement with the axiom of equipresence. More
restrictions on the constitutive relations are placed by Clausius–Duhem’s in-
equality (9.5). Introducing (9.33)–(9.36) in inequality (9.5) and applying the
chain differentiation rule to calculate
˙
f, one obtains
−ρ
_
s +
∂f
∂T
_
˙
T−ρ
∂f
∂V
:
˙
V−ρ
∂f
∂(∇T)

·
(∇T)−
1
T
q∇T−
_
∂f
∂v
I −σ
_
: V ≥ 0,
(9.37)
wherein use has been made of the continuity equation ρ˙ v = ∇ v = V : I.
On looking at inequality (9.37), we see that it is a linear expression in the
time derivatives
˙
T,
˙
V, and
·
(∇T); if it is assumed that there are body forces
and energy supply that ensure that the momentum and energy balance equa-
tions are identically satisfied, one can assign to these derivatives arbitrary
and independent values. Clearly inequality (9.37) will not hold unless the
coefficients of these derivatives are zero, which leads to the following results:
s = −
∂f
∂T
, (9.38)
∂f
∂V
= 0, (9.39)
∂f
∂(∇T)
= 0. (9.40)
248 9 Rational Thermodynamics
Relation (9.38) is classical and from the next ones, it is deduced that the free
energy f (and as a corollary the entropy s) is a function of v and T alone so
that
f = f(v, T), s = s(v, T). (9.41)
Defining the equilibrium pressure p by p = −∂f/∂v, which is justified as p
can only depend on the “equilibrium” variables v and T, and combining with
(9.38), one finds the following Gibbs equation:
df = −s dT −p dv. (9.42)
It is important to notice that, in contrast with classical irreversible thermo-
dynamics where the Gibbs relation is postulated from the outset, in RT it is
a derived result.
Furthermore, in virtue of the results (9.38)–(9.40), the Clausius–Duhem’s
inequality reduces to
_

∂f
∂v
I +σ
_
: V−
1
T
q ∇T ≥ 0. (9.43)
Introducing the viscous stress tensor σ
(v)
defined by
σ
(v)
(v, T, V, ∇T) = σ(v, T, V, ∇T) +p(v, T)I, (9.44)
(9.43) is written as
σ
(v)
: V−q
∇T
T
≥ 0, (9.45)
and represents the rate of energy dissipated per unit volume of the fluid.
Explicit expressions for the constitutive equations for q and σ
(v)
are di-
rectly obtained by using the representation theorems for isotropic tensors
(Truesdell and Toupin 1960). In the linear approximation, when second and
higher-order terms in ∇T and V are omitted, it is left as an exercise (Prob-
lem 9.7) to prove that
q = −λ(v, T)∇T, σ
(v)
= γ(v, T)(∇ v)I + 2η(v, T)V. (9.46)
One recognizes the Fourier law with λ the heat conductivity coefficient and
the Newton’s law of hydrodynamics, with η the dynamical shear viscosity
and the coefficient γ related to the bulk viscosity; of course, the scalars λ, γ,
and η are all three functions of v and T. The fluids described by the linear
constitutive relations (9.46) are usually referred to as Fourier–Stokes–Newton
fluids. After substitution of (9.46) in (9.45), it is observed that satisfaction
of Clausius–Duhem’s relation yields the following inequalities, which are well
known in fluid mechanics:
λ > 0, γ +
2
3
η > 0, η > 0. (9.47)
Summarizing, we can assert that, for the class of fluids described by constitu-
tive equations of the formφ = φ(v, T, ∇T, V) with φ ≡ f, s, q, σ, respectively:
9.5 Comments and Critical Analysis 249
1. The thermodynamic potentials f and s do not depend on ∇T and V, so
that f = f(v, T) and s = s(v, T).
2. The classical Gibbs equation (9.42) or equivalently Tds = du + p dv is
demonstrated to remain valid.
3. In the linear approximation, the constitutive equations for q and σ
(v)
are
the traditional equations of Fourier and Stokes–Newton, respectively.
4. As a side result of Clausius–Duhem’s inequality, the heat conductivity
and the viscosity coefficients are shown to be positive.
Despite the fact that rational thermodynamics radiates some taste of el-
egance and generality, it has been the subject of acrid criticisms, which are
discussed in Sect. 9.5.
9.5 Comments and Critical Analysis
The axiomatic approach of RT has been the subject of severe critical obser-
vations (e.g. Lavenda 1979; Woods 1981; Rivlin 1984) for the lack of physical
background and touch with experiments. In what follows, we shall shortly
discuss the most frequently criticisms addressed against the formalism.
9.5.1 The Clausius–Duhem’s Inequality
The basic idea is to use the Clausius–Duhem’s inequality to place restrictions
on the form of the constitutive equations. However, the original formulation
of Clausius–Duhem’s inequality is restricted to processes taking place be-
tween two equilibrium states. In RT, it is applied, without any justification,
to arbitrary processes taking place between non-equilibrium states.
Furthermore, the local Clausius–Duhem’s inequality amounts to admit
that the positiveness of the entropy production σ
s
is a necessary and suffi-
cient condition allowing to restrict the range of acceptable constitutive rela-
tions. However, it is not proved that σ
s
≥ 0 remains valid when truncated
expressions of σ
s
are used, as is generally the case in RT. By performing
a series expansion as σ
s
= σ
1
+ σ
2
+ σ
3
+ , where σ
j
(j = 1, 2, 3, . . .) is
the entropy production at the j order, it is absolutely not ensured that the
entropy production remains positive definite at any order of approximation.
9.5.2 Axiom of Phlogiston
The term phlogiston is borrowed from Woods (1981) and it designates a neg-
ative mass medium permeating all bodies and expelled by heat. The axiom
250 9 Rational Thermodynamics
concerns the property that in any process, time derivatives such as
˙
T,
·
∇T,
˙
V can be given arbitrary and independent values of T, ∇T, and V. In nor-
mal circumstances, in the balance equations of momentum and energy, the
body force F and the source term r are known and take well-specified values:
the deformation (or velocity) and temperature fields are then determined by
solving the equations after that initial and boundary conditions are given. In
RT, the procedure is reversed: for any value of the velocity v (or deformation
u) and the temperature T fields, it is admitted that one can select appropri-
ately F and r so that the equations of momentum and energy are identically
satisfied and do not impose any restriction on the set of variables. In a real
experience, F and r are specified by physical rules, which are beyond our
control and therefore they cannot be specified throughout the medium and
in that respect, it can be said that the phlogiston axiom destroys the empiri-
cal content of the balance equations of momentum and energy. The presence
of quantities F and r is required to guarantee that the state variables and
their time derivatives can be varied independently. This limits seriously the
domain of applicability of the theory as it fails to describe processes for which
the variables are not independent of their time variations.
The problems linked with the use of the phlogiston axiom can be however
circumvented thanks to an elegant technique proposed by Liu (1972) and
outlined in Appendix 1.
9.5.3 The Meaning of Temperature and Entropy
It is important to realize that in a majority of works on RT, entropy and
temperature are considered as primitive undefined objects and their physical
meaning is not a subject of deep concern. Regarding the entropy, it is simply
given by a constitutive relation expressing its functional dependence with re-
spect to the selected variables while the temperature T remains an undefined
variable which is specified by the laws it satisfies. Quoting Truesdell (1968),
“As for physical meaning, I claim no physical applicability for anything I
ever say. . . Whether a theory applies to a given piece of material at a given
time is something very important, but something that the theorist cannot be
expected to tell, in thermodynamics or any other theory.” It must however
be added that in the later developments (e.g. Day 1972; Coleman and Owen
1974; Serrin 1979; Coleman et al. 1981; Kratochvil and Silhavy 1982), more
attention has been paid to prove the existence of an entropy as a state func-
tion as well as an absolute non-equilibrium temperature. The existence of
entropy and absolute temperature is no longer postulated but deduced from
statements involving cyclic processes; nevertheless, the fundamental problem
of an unambiguous definition of temperature and entropy outside equilibrium
remains called into question. For example, some people have questioned the
measurability of the variable T introduced in the theory: a priori there should
9.5 Comments and Critical Analysis 251
be no reason to identify this T with the temperature given by a gas thermome-
ter or a thermocouple. Recent investigations (Muschik 1977; Casas-V´ azquez
and Jou 2003; Crisanti and Ritort 2003) have brought out that a precise defi-
nition of temperature and entropy is of prime importance in non-equilibrium
thermodynamics. In addition, it was proved that the entropy used in rational
thermodynamics is not unique. Meixner (1973a, 1973b) has shown that there
exists an infinity of functionals, all deserving the name of entropy, that pos-
sess the property to satisfy the Clausius–Duhem’s inequality and that there
is no criterion that permits to favour one definition over the others. Meixner’s
arguments were reinforced by a more particular analysis by Day (1977) who
demonstrated the non-uniqueness of entropy in the case of heat conduction
in a rigid isotropic body with memory. Another example of a system having
many different types of entropy was given by Coleman and Owen (1975) and
concerns elastic–plastic materials. The failure of the entropy to be not unique
is not eliminated in the state-process version of the theory.
9.5.4 Axiom of Frame-Indifference
This axiom, that requires the invariance of constitutive relations under time-
dependent accelerations and rotations of the actual reference frame, has
been the subject of intense debate and controversy. As first observed by
M¨ uller (1972) and Edelen and McLennan (1973), the axiom excludes phys-
ical processes which are Galilean invariant. Frame-indifference is not sat-
isfied in a whole series of disciplines; in particular in rational mechanics.
Newton’s law of motion formulated in a non-inertial system is “objective”
but not frame dependent as the non-inertial forces depend explicitly on the
angular velocity of the reference frame; in the kinetic theory of gases, the
Burnett equations generalizing the classical Fourier–Stokes–Newton’s laws
have also been shown to be frame dependent; in the theory of turbulence,
it is experimentally observed that the turbulent viscosity takes different val-
ues according reference frame is inertial or non-inertial (Lumley 1983). Fur-
ther violations of material indifference are found in rheology (Bird and de
Gennes 1983) and molecular hydrodynamics (Hoover et al. 1981). Another
illustration of non-respect of frame-indifference is provided by the classical
theory of irreversible processes: referring to Chap. 2, we know that the phe-
nomenological coefficients L
αβ
are depending on the angular velocity when
measured in a rotating frame. It can be said that actually, there are serious
evidences against the universality of the axiom of frame-indifference which
has to be regarded as “a convenience rather than a principle”, as concluded by
Edelen and McLennan (1973). Notwithstanding, axiom of frame-indifference
is widely adopted throughout rational thermodynamics and has disturbed
people working in continuum mechanics. Fortunately, most results of con-
tinuum mechanics are established in inertial frames so that the effects of
non-inertial forces can be ignored.
252 9 Rational Thermodynamics
9.5.5 The Entropy Flux Axiom
It is also largely admitted in rational thermodynamics that the entropy flux
is given by J
s
= q/T (plus eventually a term −µJ/T in presence of matter
diffusion) as one may expect from the Clausius–Planck’s formula. Although
this result is correct in the linear approximation, this is certainly not true
in higher-order theories as confirmed by the kinetic theory of gases (Grad
1958) or extended irreversible thermodynamics. Referring to Chap. 7, for a
wide class of processes, the entropy flux for pure substances is not the heat
flux divided by the temperature but of the more general form
J
s
=
q
T

1
3
β

(tr σ
v
)q −β

0
σ
v
q, (9.48)
wherein β

and β

are phenomenological coefficients and
0
σ
v
is the deviatoric
part of the viscous stress tensor σ
v
. This result is one of the most impor-
tant differences between rational thermodynamics and extended irreversible
thermodynamics.
9.5.6 The Axiom of Equipresence
This axiom states that all constitutive relations depend precisely on the same
set of variables, unless it is proved otherwise. As a matter of fact, the con-
clusion always shows that the response functions do not depend generally on
the whole set of variables. Of course, there is no physical argument for such
an axiom, which is essentially a mathematical convenience while formulat-
ing constitutive equations. It is certainly overweening to elevate equipresence
to the status of axiom or principle because it represents merely a technical
commodity.
To conclude, RT is an axiomatic theory characterized by generality and
mathematical elegance. Unfortunately, mathematical rigour has been ob-
tained at the detriment of physical insight and this explains some lack of
interest from some corporations of physicists and engineers. Nevertheless, it
is our opinion that it is a theory that deserves a close attention. As mentioned
earlier, some of the criticisms should be moderated and some of them are even
avoidable, for instance the phlogiston can be circumvented by appealing to
Liu’s technique while other criticisms, like these addressed again the axiom
of equipresence, are of minor consequence. It was the merit of RT to be the
first non-equilibrium formalism to get rid of the local equilibrium hypoth-
esis and to go beyond the linear regime to which is restricted the classical
theory of non-equilibrium processes. RT has been applied to a huge number
of problems mainly in the fields of non-linear elasticity, coupled mechanical,
thermal and electro-magnetic phenomena, rheology, wave propagation, and
9.5 Comments and Critical Analysis 253
shock waves. Most of them are published in the “Archive of Rational Mechan-
ics and Analysis” which is the privileged tribune of “rationalists”. Moreover,
the tools of RT have been widely applied in other formalisms as rational
extended thermodynamics (M¨ uller and Ruggeri 1998), theories with internal
variables (Maugin 1999) or GENERIC (
¨
Ottinger 2005).
Appendix 1: Liu’s Lagrange Multipliers
An elegant alternative to the admissibility axiom of rational thermodynamics,
i.e. the necessary and sufficient conditions to satisfy the Clausius–Duhem’s
inequality, was proposed by Liu (1972). He was able to show that the entropy
inequality (9.5) becomes valid for completely arbitrary variations of the vari-
ables at the condition to consider the balance equations of mass, momentum
and energy as mathematical constraints. To be more explicit, each balance
equation is multiplied by an appropriate factor, named Lagrange multiplier
by analogy with the extremization problem in mathematics, and the resulting
vanishing quantity is added to the left-hand side of the entropy inequality
(9.5). Let us illustrate the procedure by means of the simple example of heat
conduction in a rigid isotropic body without source term. The behaviour of
the temperature field is governed by one single field equation, namely the
energy balance
ρ ˙ u = −∇ q, (9.49)
it being understood that there are restrictions placed by the entropy inequality
ρ˙ s +∇
_
q
T
_
≥ 0. (9.50)
Closure relations are provided by the following set of constitutive equations:
u = u(T, ∇T), s = s(T, ∇T), q = q(T, ∇T). (9.51)
The requirement that the temperature field satisfying the entropy inequality
(9.50) must also be a solution of the energy balance equation (9.49) is inter-
preted by Liu as a constraint. It was proved by Liu that this constraint can
be eliminated by the introduction of Lagrange multipliers and by writing the
entropy inequality under the new form
ρ˙ s +∇
_
q
T
_
−Λ(ρ ˙ u +∇ q) ≥ 0; (9.52)
the quantity Λ designates the Lagrange multiplier, which depends generally
on the whole set of variables and must be determined from the formalism.
After that the constitutive relations (9.51) are inserted in inequality (9.52)
and all differentiations are performed, one obtains a relation that is linear in
the arbitrary derivatives
˙
T,
·
∇T, ∇(∇T):
254 9 Rational Thermodynamics
ρ
_
∂s
∂T
−Λ
∂u
∂T
_
˙
T +ρ
_
∂s
∂(∇T)
−Λ
∂u
∂(∇T)
_
(
·
∇T)
+
_
1
T
−Λ
_
∂q
∂(∇T)
: ∇(∇T) +
__
1
T
−Λ
_
∂q
∂T

1
T
2
q
_
∇T ≥ 0. (9.53)
Since this inequality could be violated unless the coefficients of these deriva-
tives vanish, it is found that:
∂s
∂T
−Λ
∂u
∂T
= 0,
∂s
∂(∇T)
−Λ
∂u
∂(∇T)
= 0, Λ =
1
T
. (9.54)
An important result is the third relation from which it is concluded that
the Lagrange multiplier can be identified as the inverse of the temperature.
Moreover if it is recalled that the free energy is defined as f = u − Ts, it
follows from the first and the second relation (9.54) that
∂f
∂T
= −s,
∂f
∂(∇T)
= 0. (9.55)
This implies that f is independent of ∇T and that f satisfies the classical
Gibbs relation df = −sdT. Another consequence of (9.55) is that the entropy
s is also independent of the temperature gradient; this property is also shared
by the internal energy u, in virtue of the definition of f.
There still remains from (9.53) the residual inequality
q ∇T ≤ 0. (9.56)
Clearly, the simplest way to guarantee that (9.56) is negative definite is to
assume that
q = −λ(T)∇T with λ ≥ 0, (9.57)
and we are back with Fourier’s law. From now on, the procedure is classical.
Replacing (9.57) in the energy balance equation (9.49) and writing for u(T) a
constitutive equation such that ˙ u = c
˙
T, with c the specific heat capacity, one
is led to the diffusion equation c
˙
T = ∇ (λ∇T) which, after that initial and
boundary conditions are specified, allows us to determine the temperature
distribution in the body.
Appendix 2: Rational Extended Thermodynamics
The formulation of EIT, in Chap. 7, was inspired by the concepts and methods
of CIT. But EIT may equivalently be described by making use of the tools
and structure of rational thermodynamics: it is then referred to as rational
extended thermodynamics.
As an illustration, consider the problem of hyperbolic heat conduction in
a rigid isotropic body, in absence of an energy source term. The relevant
9.5 Comments and Critical Analysis 255
variables are the internal energy u and the heat flux q. The time evolution of
u is governed by the balance law of energy (9.49) while the evolution equation
of q will be cast in the general form
ρ ˙ q = −∇ Q+σ
q
, (9.58)
where Q (a second-order tensor) denotes the flux of the heat flux and σ
q
(a
vector) is a source term. These quantities must be formulated by means of
constitutive equations which, for simplicity, will be given by
Q = Q(u, q) = a(u, q
2
)I, σ
q
= σ
q
(u, q) = b(u, q
2
)q, (9.59)
wherein the scalars a(u, q
2
) and b(u, q
2
) are unknown functions of u and q
2
to be determined.
Following Liu’s technique, the entropy inequality will be formulated in
such a way that the constraints imposed by the energy balance and (9.58)
are explicitly introduced via the Lagrange multipliers Λ
0
(u, q) (a scalar) and
Λ
1
(u, q) (a vector), so that
ρ˙ s +∇ J
s
−Λ
0
(ρ ˙ u +∇ q) −Λ
1
(ρ ˙ q +∇ Q−σ
q
) ≥ 0, (9.60)
where s(u, q
2
) and J
s
(u, q) are arbitrary functions of u and q. By differenti-
ating s and J
s
with respect to u and q, and rearranging the various terms,
one may rewrite (9.60) as:
ρ
_
∂s
∂u
−Λ
0
_
˙ u +ρ
_
2
∂s
∂q
2
q −Λ
1
_
˙ q +
∂J
s
∂u
∇u +
∂J
s
∂q
: ∇q
−Λ
0
∇ q −Λ
1
∇u
∂a
∂u
−2Λ
1
q : ∇q
∂a
∂q
2
+bΛ
1
q ≥ 0. (9.61)
Since inequality (9.61) is linear in the arbitrary derivatives ˙ u, ˙ q, ∇u, ∇q,
positiveness of (9.61) requires that their respective factors vanish, from which
it results that
∂s
∂u
= Λ
0
(≡ θ
−1
), 2
∂s
∂q
2
q = Λ
1
(≡ γ(u, q
2
)q) (9.62)
and
∂J
s
∂u
=
∂a
∂u
Λ
1
,
∂J
s
∂q
= Λ
0
I + 2Λ
1
q
∂a
∂q
2
, (9.63)
wherein we have identified Λ
0
with θ
−1
, the inverse of a non-equilibrium
temperature (see Box 7.4) and, without loss of generality, Λ
1
with γ(u, q
2
)q
where γ is an arbitrary function of u and q
2
.
Taking into account of the results (9.62) and (9.63), the entropy inequality
(9.61) reduces to the residual inequality

1
q ≥ 0. (9.64)
256 9 Rational Thermodynamics
The fact that the entropy flux is an isotropic function implies that
J
s
= ϕ(u, q
2
)q, (9.65)
which, substituted in (9.63), yields
∂ϕ
∂u
= γ
∂a
∂u
(a) and (ϕ −θ
−1
)I + 2
_
∂ϕ
∂q
2
−γ
∂a
∂q
2
_
qq = 0 (b). (9.66)
Since the dyadic product qq is generally not zero, it follows that
∂ϕ
∂q
2
= γ
∂a
∂q
2
(a) and ϕ = θ
−1
(b). (9.67)
The second result (9.67) is important as it indicates that ϕ is equal to the
inverse of the temperature and as a consequence that the entropy flux (9.65)
is given by the usual expression
J
s
= q/θ. (9.68)
Moreover, in virtue of (9.66a) and (9.66b), one has

−1
≡ dϕ =
∂ϕ
∂u
du +
∂ϕ
∂q
2
dq
2
= γda, (9.69)
a result that will be exploited to obtain the final expression of the evolution
equation of the heat flux. Indeed, after making use of the results σ
q
= bq
and ∇ Q = ∇a = γ
−1
∇θ
−1
in (9.58), it is found that
ρ ˙ q = −
1
γ
∇θ
−1
+bq. (9.70)
Dividing both members of (9.70) by b and setting ρ/b = −τ, (γb)
−1
= L, one
recovers a Cattaneo-type relation
τ ˙ q = L∇θ
−1
−q, (9.71)
wherein τ can be identified as a relaxation time and L as a generalized heat
conductivity. In terms of these coefficients, the Lagrange multiplier Λ
1
is
given by
Λ
1
= γq = −(τ/ρL)q. (9.72)
Let us finally derive the Gibbs equation, it follows directly from (9.62) that
ds =
∂s
∂u
du +
∂s
∂q
2
dq
2
=
∂s
∂u
du +Λ
1
dq = θ
−1
du −
_
τ
ρL
_
q dq, (9.73)
where the identification (9.72) has been introduced. It is worth to stress that
this relation is similar to (7.24). Equality of the second-order derivatives
9.6 Problems 257
of s in (9.73) leads to the Maxwell relation ∂θ
−1
/∂q
2
=
1
2
(∂γ/∂u) and, after
integration,
θ
−1
(u, q
2
) = T
−1
(u) +
1
2
_
∂γ(u, q
2
)
∂u
dq
2
, (9.74)
where T(u) is the local equilibrium temperature. Clearly, the non-equilibrium
temperature reduces to its equilibrium value when the factor γ (or equiva-
lently the Lagrange multiplier Λ
1
) is independent of u.
We close this analysis with some considerations about the sign of the
various coefficients. Stability of (local) equilibrium requires that the second
variation of entropy with respect to the state variables u and q is negative
definite, so that in particular,

2
s
∂q
2
= γ ≤ 0, (9.75)
but, referring to the entropy inequality (9.64) which can be cast in the form
bγq
2
≥ 0, it follows that b < 0 whence L ≡ (bγ)
−1
≥ 0 and τ = −(ρ/b) ≥ 0.
These results confirm the positive definite property of the heat conductivity
coefficient and the relaxation time.
As compared with the results of Appendix 1, it is seen that in the present
description, the non-equilibrium entropy (9.73) depends on the heat flux. This
is due to the fact that q is assumed to have an evolution equation (9.58) of
its own, leading to the introduction of a supplementary Lagrange multiplier
in equation (9.60); instead, in Appendix 1 the temperature gradient was not
assumed to be described by a proper evolution equation so that no extra
Lagrange multiplier was needed.
9.6 Problems
9.1. Dissipated energy. Verify that the left-hand side of the Clausius–Duhem’s
inequality (9.5) can be identified with the rate of dissipated energy Tσ
s
at
the condition that the corresponding entropy flux is given by J
s
= q/T.
9.2. Generalized Fourier’s law. Show that the generalized Fourier law as the
time derivative of the heat flux q(t) =
_
t
−∞
λ(t −t

)∇T(t

)dt

is equal to
Cattaneo’s relation ˙ q = −q/τ − (λ/τ)∇T when the memory kernel is given
by the expression λ(t −t

) = −(λ/τ) exp[−(t −t

)/τ], where λ is the heat
conductivity and τ is a constant relaxation time.
9.3. Objectivity. Show that the skew-symmetric (antisymmetric) part of the
velocity gradient tensor W =
1
2
[(∇v) −(∇v)
T
] transforms as W

= Q W
Q
T
+
˙
Q Q
T
under the Euclidean transformation (9.8).
258 9 Rational Thermodynamics
9.4. Objectivity. Prove that (a) ∇T and the displacement vector u of elas-
ticity are objective vectors; (b) the symmetric part of the velocity gradient
V =
1
2
[(∇v) + (∇v)
T
] and the gradient of an objective vector are objective
tensors; and (c) the deformation gradient F and the material time derivative
of an objective vector, say the heat flux q, are not objective.
9.5. Clausius–Duhem’s inequality. Establish the Clausius–Duhem’s inequal-
ity, respectively, in the case of small elastic deformations (see (9.18)) and
large deformations (refer to Box 9.1).
9.6. Large elastic deformations. Consider a material body defined by the
set of variables T, ∇T, F, and
˙
F where T is the temperature and F the
deformation tensor. Determine the restrictions placed on the constitutive
equations of f (free energy), s (entropy), S (Piola stress tensor), and Q
(heat flux vector).
9.7. Isotropic tensors. Referring to the theorems of representation of isotropic
tensors (e.g. Truesdell and Toupin 1960), the constitutive equation of the
viscous stress tensor of a heat conducting viscous fluid reads as
σ
v
= α(ρ, T)I +γ(ρ, T)(∇ v)I + 2η(ρ, T)V,
when the second- and higher-order terms in V are omitted. (a) Show that α
is zero, as a result of the property that the viscous stress tensor has to be
zero at equilibrium. (b) Decomposing the symmetric velocity gradient tensor
in a bulk and a deviatoric part V =
1
3
(∇ v)I +
0
V, with
0
V the traceless
deviator, show that
σ
v
= ζ(ρ, T)(∇ v)I + 2η(ρ, T)
0
V,
with ζ = γ +
2
3
η designating the bulk viscosity. (c) Verify that ζ > 0, η > 0.
9.8. Parabolic heat conduction. Consider a one-dimensional rigid heat con-
ductor defined by the following constitutive equations:
f = f(T, ∂T/∂x), s = s(T, ∂T/∂x), q = q(T, ∂T/∂x).
and the energy balance equation
ρ∂u/∂t = ∂q/∂x +ρr.
Determine the resulting heat conduction equation for the temperature field,
given by
ρT

2
f
∂T
2
∂T
∂t
= −
∂q
∂T
∂T
∂x

∂q
∂(∂T/∂x)

2
T
∂x
2
+ρr.
Show that this equation is of the parabolic type if ∂
2
f/∂T
2
> 0 and ∂q/∂T <
0 so that no propagation of wave can occur.
9.6 Problems 259
9.9. Hyperbolic heat conduction. Reconsider Problem 9.8 but with T, ∂T/∂t,
∂T/∂x as variables. Show that in the present case, the temperature equation
is of the hyperbolic type.
9.10. Viscous heat conducting fluid. A viscous heat conducting fluid is char-
acterized by the following constitutive equation φ = φ(v, T, ∇v, ∇T, V) with
φ ≡ f, s, q, σ. Determine the restrictions placed by Clausius–Duhem’s in-
equality on the constitutive equations.
9.11. Thermodiffusion. Study the problem of thermodiffusion in a binary
mixture of non-viscous fluids within the framework of rational thermody-
namics.
9.12. Liu’s technique. Reformulate the problem of a heat conducting viscous
fluid (see Sect. 9.4) by using Liu’s technique.
Chapter 10
Hamiltonian Formalisms
A Mathematical Structure of Reversible
and Irreversible Dynamics
In the theories discussed so far, the limitations on the form of the constitutive
equations arise essentially from the application of the second law of thermo-
dynamics. However, no restrictions were placed on the reversible terms of the
time evolution equations, as they do not contribute to the entropy production.
Such terms may be either gyroscopic forces, as Coriolis force, or convected
time derivatives, as Maxwell or corotational derivative, which are of frequent
use in rheology. It is shown here that requirement of a Hamiltonian structure
leads to restrictions on the reversible part of the evolution equations.
Hamiltonian formulations have always played a central role not only in
mechanics but also in thermodynamics. They have been identified at dif-
ferent levels of description: from the microscopic one (classical mechanics,
kinetic theory) to the macroscopic one (theory of elasticity, frictionless fluid
flows, equilibrium thermodynamics). Despite their theoretical appeal, and
except rare attempts, Hamiltonian methods have not been fully exploited
in presence of irreversible effects. The reasons that militate in favour of a
Hamiltonian description are numerous: the first one is conciseness, as the
whole set of balance equations are expressed in terms of a limited number of
potentials, generally, one single potential is sufficient. From a practical point
of view, there exist, in parallel, powerful numerical methods, which have
been developed for Hamiltonian systems to obtain approximate solutions, as
the task of scientists is not only to derive equations but also to find solu-
tions. Moreover, besides their power of synthesis and their practical interest,
Hamiltonian descriptions are also helpful for the physical interpretation of
the processes: indeed, the generating potentials may generally be identified
with well-defined physical quantities as the mechanical energy, the energy
of deformation, the thermodynamic potentials, etc. Hamiltonian techniques
have also a wide domain of applicability, as they are not restricted to the
linear regime. Finally, they place several restrictions on the possible forms of
the constitutive equations, complementing those provided by the second law
of thermodynamics.
261
262 10 Hamiltonian Formalisms
The main part of this chapter will be devoted to the presentation of a
general equation for the non-equilibrium reversible–irreversible coupling, ab-
breviated as GENERIC, which consists in a generalization of the Poisson
bracket formalism originally proposed in the framework of classical mechan-
ics (Grmela and
¨
Ottinger 1997;
¨
Ottinger and Grmela 1997;
¨
Ottinger 2005).
The principal motivation for developing GENERIC is the modelling of the
flow properties of rheological fluids; this is achieved by formulating general
time evolution equations taking the same universal form whatever the nature
of the state variables. Another particularity of GENERIC is that the evo-
lution equations are expressed in terms of two appropriate thermodynamic
potentials, called generators, taking the form of the total energy and a dissi-
pation potential. GENERIC can be viewed as an extension of both Hamilton’s
equations and Landau’s potential (Landau 1965). Before examining the main
tenets of GENERIC and illustrating its application by a number of simple
examples, let us preliminarily recall the Hamiltonian description of classical
mechanics.
10.1 Classical Mechanics
Consider a collection of N particles characterized by the set of variables
x = (r, p) with r = (r
1
, . . . , r
N
) and p = (p
1
, . . . , p
N
) denoting the position
and momentum vectors of the particles. Denoting by E = K + V the total
energy (kinetic and potential energies) or the Hamiltonian of the system, it
is well known that the time evolution of the set of variables x is given by the
Hamilton equations
dx
dt
=
ˆ
L
∂E
∂x
, (10.1)
with
ˆ
L =
_
0 1
−1 0
_
,
or, more explicitly,
dr
dt
=
∂E
∂p
,
dp
dt
= −
∂E
∂r
. (10.2)
The Poisson matrix
ˆ
L expresses the reversible kinematics of x; reversibility
means invariance with respect to the change t →−t, it being understood that
concomitantly r, p → r, −p and E(r, p) → E(r, −p). Instead of working
with the matrix
ˆ
L, an equivalent way to formulate the dynamical equations
(10.2) is to introduce the Poisson bracket
¦A, B¦ =
∂A
∂r
∂B
∂p

∂A
∂p
∂B
∂r
, (10.3)
10.1 Classical Mechanics 263
which is related to
ˆ
L by
¦A, B¦ =
_
∂A
∂x
,
ˆ
L
∂B
∂x
_
(10.4)
with (,) denoting the scalar product, whereas A and B are regular functions
of x. The evolution of x is now governed by
dA
dt
= ¦A, E¦, (10.5)
where A(x) is an arbitrary function of x, not dependent explicitly on time.
This is directly seen by writing explicitly both members of (10.5), i.e.
∂A
∂r
dr
dt
+
∂A
∂p
dp
dt
=
∂A
∂r
∂E
∂p

∂A
∂p
∂E
∂r
, (10.6)
and after identifying the coefficients of ∂A/∂r and ∂A/∂p in both sides of
(10.6), one recovers indeed the evolution equations (10.2). An illustrative
example is presented in Box 10.1.
Box 10.1 A Classical Mechanics Illustration: The Harmonic
Oscillator
Consider a particle of constant mass m fixed at the end of a spring of
stiffness H. The problem is assumed to be one dimensional with x denoting
the position of the particle and p its momentum. The total energy is
E(x, p) =
1
2
p
2
m
+
1
2
Hx
2
. (10.1.1)
In virtue of (10.2), the evolution equations of x and p are given by
dx
dt
=
p
m
,
dp
dt
= −Hx. (10.1.2)
After elimination of p, one recovers Newton’s law
m
d
2
x
dt
2
= −Hx. (10.1.3)
Making use of the definition (10.4) of the Poisson bracket ¦A, B¦ one has
¦A, B¦ =
_
∂A
∂x
∂A
∂p
__
0 1
−1 0
_




∂B
∂x
∂B
∂p




, (10.1.4)
which is in agreement with the result (10.3). Accordingly, the Poisson
bracket ¦A, E¦ describing the harmonic oscillator is given by
¦A, E¦ =
∂A
∂x
p
m

∂A
∂p
Hx (10.1.5)
264 10 Hamiltonian Formalisms
and the corresponding evolution equation (10.5) takes the form
∂A
∂x
dx
dt
+
∂A
∂p
dp
dt
=
∂A
∂x
p
m

∂A
∂p
Hx. (10.1.6)
Comparison of the coefficients of ∂A/∂x and ∂A/∂p in both sides of (10.1.6)
gives back the equations of motion (10.1.2).
Expression (10.4) is a Poisson bracket if the two following conditions are
fulfilled:
(1)
¦A, B¦ = −¦B, A¦ (antisymmetry), (10.7)
(2)
¦A, ¦B, C¦¦ +¦B, ¦C, A¦ +¦C, ¦A, B¦¦ = 0 ( Jacobi’s identity). (10.8)
The antisymmetry property of the Poisson bracket implies the antisymme-
try of operator
ˆ
L; the Jacobi’s identity imposes additional severe restrictions
and expresses the time structure invariance of the Poisson bracket ¦A, B¦.
It is precisely this identity what gives information on the reversible contri-
butions to the dynamics (for instance, non-linear convective contributions
to the time derivatives), which is not available from the positiveness of the
entropy production.
Evolution equations as (10.1) or its equivalent (10.5) are reversible time
evolution equations as they are invariant with respect to time reversal t →−t,
a result well known in classical mechanics. In more general situations as
these encountered in continuum physics, dx/dt is the sum of a reversible
and a non-reversible contribution and we will see in Sect. 10.2 how to inte-
grate irreversible dynamics in the framework of GENERIC. In mathematical
terminology, the manifold of the phase space, a Hamiltonian, and an antisym-
metric matrix relating the time derivative of the variables with the partial
derivatives of the Hamiltonian with respect to its variables are known as a
symplectic manifold.
10.2 Formulation of GENERIC
GENERIC may be considered as an extension of Landau’s idea accordingly
the time evolution of a state variable x, like mass density or energy, towards
its equilibrium value x
eq
is governed by the relation
dx
dt
= −
δΨ
δx
, (10.9)
10.2 Formulation of GENERIC 265
where Ψ is a given potential which is minimum in the equilibrium or in the
steady state:
δΨ
δx
= 0 at x = x
eq
, (10.10)
with δ/δx denoting the functional or Volterra derivative with respect to the
variable x; if Ψ =
_
ψ dV is a simple scalar function of the variables, then
δΨ/δx reduces to the usual partial derivative δΨ/δx = ∂ψ/∂x; the situation
is more complicated when Ψ depends in addition on the gradient of x, if x
is assumed to be a scalar, then
δΨ
δx
=
∂ψ
∂x
−∇
_
∂Ψ
∂(∇x)
_
. (10.11)
However, a relaxation-type equation as (10.9) describing the irreversible ap-
proach to equilibrium is too restrictive and cannot pretend to cope with
general processes of continuum physics.
When both reversible and irreversible processes are present, one formu-
lates, in the framework of GENERIC, a general time evolution equation in
which the evolution of a variable x is expressed in terms of two potentials, the
total energy E of the overall system and a dissipation function Ψ: explicitly,
one has
dx
dt
=
ˆ
L
δE
δx
+
δΨ
δ(δS/δx)
. (10.12)
The quantity S has the physical meaning of the entropy of the overall system.
The dissipation potential Ψ, which is a real-valued function of the derivatives
δS/δx, possesses the following properties: Ψ(0) = 0, it is minimum at x =0
and is convex in the neighbourhood of 0. Note that in the GENERIC frame-
work, it is generally assumed that the overall system is isolated from its
environment so that
dE
dt
= 0,
dS
dt
> 0. (10.13)
Notice also that in the particular case that Ψ is a quadratic function of δS/δx,
of the form
1
2
(∂S/∂x)
ˆ
M (∂S/∂x), (10.12) takes the more familiar form
dx
dt
=
ˆ
L
δE
δx
+
ˆ
M
δS
δx
. (10.14)
The matrices
ˆ
L and
ˆ
M, operating on the functional derivatives of E and S,
produce the reversible and irreversible contributions to the evolution of x,
respectively. These matrices must satisfy some conditions:
ˆ
L must be an-
tisymmetric and verify Jacobi’s identity, and
ˆ
M must be symmetric and
positive-definite to ensure the positiveness of the dissipation rate. However,
the restrictions on
ˆ
L and
ˆ
M mentioned above are not sufficient to guarantee a
thermodynamically consistent description of the dynamics of the system, and
266 10 Hamiltonian Formalisms
two supplementary restrictions, called degeneracy conditions, are introduced,
namely
ˆ
L
δS
δx
= 0, (10.15)
ˆ
M
δE
δx
= 0. (10.16)
The first condition expresses that entropy cannot contribute to the reversible
nature of
ˆ
L and therefore is not modified by the reversible part of the dy-
namics; the second condition implies the conservation of total energy by the
dissipative contribution to the dynamics. Relations (10.12) or (10.14) repre-
sent the GENERIC extension of the Landau equation (10.9) and express the
universal structure of non-equilibrium thermodynamics, which is completely
specified by the knowledge of the four quantities E, S (or Ψ),
ˆ
L, and
ˆ
M.
By analogy with classical mechanics where the Hamilton equations (10.1)
can be replaced by the Poisson brackets (10.5), it is equivalent to write (10.14)
in the form
dA
dt
= ¦A, E¦ + [A, S], (10.17)
where
¦A, E¦ =
_
δA
δx
,
ˆ
L
δE
δx
_
(10.18)
is a Poisson bracket with the antisymmetry property ¦A, E¦ = −¦E, A¦,
whereas
[A, S] =
_
δA
δx
,
ˆ
M
δS
δx
_
(10.19)
is the so-called Landau symmetric bracket, i.e. [A, S] = [S, A] satisfying in
addition the positiveness property [S, S] > 0. In terms of the dissipation
potential Ψ, the above bracket will be given the form
[A, S] =
_
δA
δx
,
δΨ
δ(δS/δx)
_
, (10.20)
which is a generalization of (10.19). In the foregoing, we will illustrate the
use of GENERIC with two examples: isothermal hydrodynamics and matter
diffusion in a binary moving mixture. Several applications to polymer solu-
tions have been worked out, but we refer to
¨
Ottinger and Grmela (1997) and
¨
Ottinger (2005) for an overview of this topic.
10.2.1 Classical Navier–Stokes’ Hydrodynamics
Let us consider the motion of a compressible one-component viscous fluid
assumed to take place under isothermal conditions. The following four steps
govern the construction of the GENERIC formalism:
10.2 Formulation of GENERIC 267
Step 1. Selection of the state variables
Just like in other thermodynamic theories, the selection of state variables is
subordinated by the nature of the process under consideration and the degree
of accuracy that one wishes to achieve. For the present problem, the set x of
variables is conveniently selected as:
x : ρ(mass density), u(momentum). (10.21)
In GENERIC, it is more convenient to work with the momentum rather than
with the velocity field v(= u/ρ). The pressure p is not included among the
variables because it will be expressed in terms of the independent fields by
means of a constitutive equation.
Step 2. Thermodynamic potential
Since the temperature T is fixed and uniform, the Helmholtz free energy is
the potential that will play the central role, it is expressed by
Φ(ρ, u) =
_
φ(ρ, u)dV = E(ρ, u) −TS(ρ), (10.22)
with the total energy and the total entropy given by
E(ρ, u) =
_
[(u u)(2ρ)
−1
+ε(ρ)]dV, (10.23a)
S(ρ) =
_
s(ρ)dV , (10.23b)
where ε and s denote, respectively, the internal energy and the entropy re-
ferred per unit volume; just like ε, the entropy s cannot depend on the
momentum u as it is a specific thermodynamic quantity. Here we use the
notation ε instead of u to designate the internal energy, thus avoiding the
confusion with the momentum u.
For further purpose, it is interesting to observe that φ
u
α
= u
α
/ρ is the
velocity field; the notation φ
u
α
standing for ∂φ/∂u
α
where ∂/∂u
α
is the
partial derivative with respect to the component u
α
of the momentum.
Step 3. Hamiltonian reversible dynamics
Our purpose is to specify the reversible contribution to the evolution equa-
tions of hydrodynamics. Let A =
_
a dV denote a regular function of ρ and
u whose time evolution is given by
dA
dt
=
_
(a
ρ

t
ρ +a
u
α

t
u
α
)dV, (10.24)
where a
ρ
= ∂a/∂ρ and a
u
α
= ∂a/∂u
α
.
Following the lines of thought of GENERIC, the time evolution of A can
be cast in the form
dA
dt
= ¦A, Φ¦, (10.25)
268 10 Hamiltonian Formalisms
where ¦A, Φ¦ is a Poisson bracket which, in the present problem, is defined as
¦A, Φ¦ =
_
¦ρ[(∂
α
a
ρ

u
α
−(∂
α
φ
ρ
)a
u
α
] +u
γ
[(∂
α
a
u
γ

u
γ
−(∂
α
φ
u
γ
)a
u
α
]¦dV
(10.26)
with ∂
α
denoting the spatial derivative with respect to the x
α
coordinate,
we have also used the summation convention on repeated indices. A system-
atic construction of the expression of the Poisson bracket (10.26) has been
developed on general arguments based on the group theory by Grmela and
¨
Ottinger (1997) and
¨
Ottinger (2005). Substituting (10.24) and (10.26) in
(10.25), one obtains
_
[a
ρ
(∂
t
ρ) +a
u
α
(∂
t
u
α
)]dV
=
_
_
a
ρ
[−∂
γ
(ρφ
u
γ
)] −a
u
α
[∂
γ
(u
α
φ
u
γ
) +ρ∂
α
φ
ρ
+u
γ

α
u
γ
]
_
dV, (10.27)
after that the right-hand side of (10.26) has been integrated by parts and
the boundary conditions have been selected to make all the integrals over the
boundary equal to zero.
By identification of the coefficients of a
ρ
and a
u
α
, it is found that

t
ρ = −∂
γ
(ρφ
u
γ
), (10.28)

t
u
α
= −∂
γ
(u
α
φ
u
γ
) −ρ∂
α
φ
ρ
−u
γ

α
φ
u
γ
. (10.29)
To recover the Euler equations of hydrodynamics, one has to identify the two
last terms of the right-hand side of (10.29) with the pressure gradient, i.e.

α
p = ρ∂
α
φ
ρ
+u
γ

α
φ
u
γ
. (10.30)
Moreover, from the chain differentiation rule, one obtains

α
φ = φ
ρ

α
ρ +φ
u
γ

α
u
γ
,
which, coupled to (10.30) leads to
p = −φ +ρφ
ρ
+u
γ
φ
u
γ
= −ε +Ts +ρµ. (10.31)
This is the well-known Euler equation of equilibrium thermodynamics after
that φ
ρ
= µ has been identified with the chemical potential. As shown earlier,
one has φ
u
α
= v
α
so that (10.28) and (10.29) take the form

t
ρ = −∂
γ
(ρv
γ
), (10.32)

t
u
α
= −∂
γ
(pδ
αγ
+u
α
u
γ
ρ
−1
), (10.33)
which are the equations describing the behaviour of an Euler fluid in hydro-
dynamics.
10.2 Formulation of GENERIC 269
When the time evolution of the system is described by the Euler equations
(10.32) and (10.33), it is checked that

dt
= 0, (10.34)
which follows directly from (10.25) as a consequence of the antisymmetry
property of the Poisson bracket. It is also verified that the relations (10.32)
and (10.33) are invariant with respect to the time reversal t → −t and this
justifies the denomination “reversible” or “non-dissipative”.
Step 4. Irreversible dynamics
In the presence of dissipation, the free energy Φ will diminish in the course
of time

dt
< 0 (10.35)
and the general time evolution equation is
dA
dt
= ¦A, Φ¦ −[A, Φ], (10.36)
where, referring to (10.20), the dissipative bracket is defined as
[A, Φ] = [δA/δu
α
, δΨ/δφ
u
α
] =
_
a
u
α
[(∂ψ/∂φ
u
α
) −∂
γ
(∂ψ/∂(∂
γ
φ
u
α
))] dV
(10.37)
with Ψ =
_
ψ dV the dissipation potential. Two remarks are in form about
the formulation of (10.36) and (10.37). First, when compared to (10.17), the
change of sign in front of the dissipative bracket is a consequence of the use
of the free energy Φ instead of the entropy S as basic function; second, there
is no contribution of the mass density to (10.37) because ρ is a conserved
quantity.
To reproduce the classical Navier–Stokes’ relation, select ψ as given by the
quadratic form
ψ =
1
4
(∂
γ
φ
u
a
+∂
α
φ
u
γ
)η(∂
γ
φ
u
α
+∂
α
φ
u
γ
) > 0, (10.38)
wherein η is a phenomenological coefficient, the dynamic viscosity, which is a
positive quantity to guarantee the positive definite property of the dissipation
potential. It is directly checked that the integrant of (10.37) is
−a
u
α

γ
_
η(∂
α
φ
u
γ
+∂
γ
φ
u
α
)
¸
. (10.39)
Following the same procedure as in Step 3, one obtains the next evolution
equations for ρ and u
α
:

t
ρ = −∂
γ
(ρφ
u
γ
), (10.40)

t
u
α
= −∂
γ
(u
α
φ
u
α
) −∂
α
p −∂
γ
P
αγ
, (10.41)
270 10 Hamiltonian Formalisms
where the symmetric pressure tensor P
αγ
is expressed by
P
αγ
= −
1
2
η(∂
α
φ
u
γ
+∂
γ
φ
u
α
) = −
1
2
η(∂
α
v
γ
+∂
γ
v
α
). (10.42)
The last term in (10.41) represents the irreversible contribution to the mo-
mentum equation as it is not invariant to the time reversal. From the iden-
tification of (10.42), it is seen that the dissipation potential (10.38) can be
cast in the form −P
v
: V, which is (2.68) for the viscous dissipation found
in the classical formulation of hydrodynamics. By incorporating in the rela-
tion (10.38) of the dissipation potential an extra term proportional to the
square of ∂
γ
φ
u
γ
, one would recover Stokes’ law between the bulk viscous
pressure and the divergence of the velocity. External forces, like gravity, can
also be incorporated by simply adding the corresponding potential, namely

_
ρg x dV to the total energy given by (10.23a).
The above results are important because they exhibit the GENERIC struc-
ture of the familiar Navier–Stokes’ equations. Indeed, accordingly to the gen-
eral rules of GENERIC, the evolution equations of hydrodynamics can be
formulated in terms of two potentials: the Helmholtz free energy and the
dissipation potential; moreover, there exists an acute separation between re-
versible and irreversible contributions. It is also important to mention some
properties of the solutions of (10.40) and (10.41): first, when the system is left
during a sufficiently long time outside the influence of external constraints, it
will tend to an equilibrium state characterized by φ
u
α
= 0, φ
ρ
= 0 for t →∞.
Second, if the potential Φ is convex (Φ ≥ 0), and because dΦ/dt < 0, Φ plays
the role of a Lyapounov function ensuring the stability of equilibrium.
It was stated in Sect. 10.1 that an equivalent description of GENERIC
consists in expressing the evolution equations in terms of the two potentials
E and S and the operators
ˆ
L and
ˆ
M. However, this kind of approach re-
quires some expertise, technique and feeling, which is outside the scope of
the present introductory monograph and therefore we refer the reader to
specialized works (Grmela and
¨
Ottinger 1997;
¨
Ottinger and Grmela 1997;
¨
Ottinger 2005).
In the present section, we have considered the flow of a one-component
fluid. In Sect. 10.2.2, we will examine the more general situation of the motion
of a binary mixture to exhibit the coupling between matter diffusion and
momentum transport.
10.2.2 Fickian Diffusion in Binary Mixtures
We consider an isotropic mixture of two Newtonian viscous fluids perfectly
miscible and chemically inert, the whole system is in motion in absence of
external forces and its temperature is assumed to be uniform. Let ρ
1
and ρ
2
be the mass densities of the two components and u
1
and u
2
their barycentric
10.2 Formulation of GENERIC 271
momenta, the average barycentric momentum is given by u = u
1
+u
2
. The
presentation of Sect. 10.2.1 will be traced back and first applied to Fickian
diffusion.
Step 1. State variables
The space of basic variables for the present problem is formed by the set
x : u, ρ
1
, ρ
2
. However, for practical reasons, it is more convenient to work
with the following equivalent set
x : u, ρ, c, (10.43)
where ρ = ρ
1

2
is the total mass density and c = ρ
1
/ρ is the mass fraction
of one of the constituents, say component 1 (note that c + c
2
= 1, with
c
2
= ρ
2
/ρ).
Step 2. Thermodynamic potential
We will use the same Helmholtz potential as in Sect. 10.2.1, i.e.
Φ(ρ, c, u) = E(ρ, c, u) −TS(ρ, c) (10.44)
with E =
_
[u u/(2ρ) +ε(ρ, c)]dV where ε is the total internal energy per
unit volume of the binary mixture. At this level of description, we have ne-
glected the contribution from the diffusion kinetic energy, which is justified
in diluted systems.
Step 3. Reversible dynamics
The reversible dynamics is written in the form dA/dt = ¦A, Φ¦ where the
Poisson bracket is now given by (Grmela et al. 1998):
¦A, Φ¦ =
_
_
ρ[(∂
α
a
ρ

u
α
−(∂
α
φ
ρ
)a
u
α
] +u
γ
[(∂
α
a
u
γ

u
α
−(∂
α
φ
u
γ
)a
u
α
] −(∂
α
c)(a
c
φ
u
α
−φ
c
a
u
α
)
_
dV. (10.45)
This is the same relation as (10.26) to which we have added a third-term
expressing the contribution from the mass fraction c. The resulting evolution
equations are obtained as before, i.e. by writing explicitly dA/dt in terms of
the variables, by integrating (10.45) by parts and by identifying the coeffi-
cients of a
ρ
, a
u
α
, and a
c
, respectively. These operations lead to the following
reversible part of the evolution equations

t
ρ = −∂
γ
(ρφ
u
γ
), (10.46)

t
u
α
= −∂
γ
(u
α
φ
u
γ
) −∂
α
p, (10.47)

t
c = −(∂
γ
c)φ
u
γ
, (10.48)
where the thermostatic pressure is still given by p = −φ + ρφ
ρ
+ u
γ
φ
u
γ
. It
is easily checked that the above equations are invariant with respect to time
reversal.
272 10 Hamiltonian Formalisms
Step 4. Irreversible dynamics
To recover Fick’s law of diffusion, it is necessary to introduce the dissipa-
tion potential. The latter is a generalization of the potential used in (10.38),
written here as ψ (10.38), with a supplementary term,
ψ = ψ(10.38) +
1
2
(∂
γ
φ
c
)D

(∂
γ
φ
c
). (10.49)
The contribution coming from diffusion is the last term at the right-hand side
of (10.49), D

is a positive phenomenological coefficient allowed to depend
on ρ and c. After that dissipation has been included, the evolution equations
(10.46)–(10.48) take the final form

t
ρ = −∂
γ
(ρv
γ
), (10.50)

t
u
α
= −∂
γ
(u
α
v
γ
) −∂
α
p −∂
γ
P
αγ
, (10.51)

t
c = −v
γ

γ
c +∂
γ
(D


γ
φ
c
), (10.52)
wherein the two first relations are the classical continuity and momentum
equations whereas the last one is the diffusion equation; the irreversible con-
tributions are the last term in (10.51), with P
αγ
still given by (10.42), and the
last term of (10.52). Denoting by J
γ
, the γ component of the flux of matter,
it follows from (10.52) that
J
γ
= −D


γ
φ
c
. (10.53)
Assuming that φ is a quadratic function of c with φ
c
= D
1
c, where D
1
is a
phenomenological constant, one recovers Fick’s law
J
γ
= −ρD∂
γ
c, (10.54)
at the condition to define the coefficient of diffusion D as D
1
D

/ρ; observe
that in contrast to D
1
, the factor D is not necessarily a constant.
One of the merits of GENERIC is to provide a systematic way to derive
the evolution equations whatever the nature and the number of basic vari-
ables. Of course, to obtain explicit relations, we need constitutive relations
for the potentials φ and ψ, which will be determined by the physico-chemical
properties of the system under study.
Although the above analysis gives an overall description of diffusion in
a two-constituent mixture, it cannot pretend to describe the behaviour of
each individual component; in particular, one has no information about the
time evolution of the particular velocities u
1
and u
2
. This would require the
introduction of u
1
and u
2
as independent variables, instead of the single
barycentric momentum u. Such a task will be achieved in Sect. 10.2.3.
10.2 Formulation of GENERIC 273
10.2.3 Non-Fickian Diffusion in Binary Mixtures
Our objective is twofold: to derive the evolution equations of the two indi-
vidual components, and to generalize the Fick’s law to non-steady situations.
For the sake of simplicity, we assume that the barycentric momentum u is
equal to zero; this assumption implies that the total variation of the total
mass density ρ is negligible in the course of time. In addition, we assume
that both fluids are non-viscous. A more general analysis, wherein the overall
velocity does not vanish and where viscosity is not neglected, can be found
in Grmela et al. (2003).
1. State variables
Here we make the one-to-one transformation
ρ
1
, ρ
2
, u
1
, u
2
→ρ, c, u, w, (10.55)
where as above ρ designates the total mass density, c is the mass fraction
of component 1, u is the barycentric momentum while the vector w stands
for
w = ρ
−1

2
u
1
−ρ
1
u
2
). (10.56)
As stated above, from now on, one supposes that u = 0 (no convection).
Once c, ρ, and w are determined, one obtains directly ρ
1
, ρ
2
, u
1
and u
2
from the inverse transformation ρ
1
= cρ, ρ
2
= (1 −c)ρ, u
1
= −u
2
= w.
2. Thermodynamic potential
The latter is similar to (10.44) but the specific kinetic energy u
2
1
(2ρ
1
)
−1
+
u
2
2
(2ρ
2
)
−1
is now equal to the diffusion kinetic energy w
2
/2ρc(1−c) instead
of the barycentric kinetic energy u
2
/2ρ, as in Sect. 10.2.2.
3. Hamiltonian reversible dynamics
The Hamiltonian setting is equivalent to that proposed in the previous
sections; inspired by (10.26) and (10.45), we will write
¦A, Φ¦ =
_
¦ρc(1 −c)[∂
α
(a
c
/ρ)φ
w
α
−∂
α

c
/ρ)a
w
α
]
−(1 −c)w
γ
[(∂
α
a
w
γ

w
α
−(∂
α
φ
w
γ
)a
w
α
]
−w
γ
[∂
α
(ca
w
γ

w
α
−∂
α
(cφ
w
γ
)a
w
α
]
_
dV. (10.57)
It follows that the non-dissipative contribution to the evolution equations
will be

t
ρ = 0, (10.58)

t
c = −∂
γ
[ρc(1 −c)φ
w
γ
]/ρ, (10.59)

t
w
α
= −ρc(1 −c)∂
α

c
/ρ) +c∂
γ
(w
α
φ
w
γ
)
−∂
γ
[(1 −c)w
α
φ
w
γ
] −(1 −c)w
γ
(∂
α
φ
w
γ
) +w
γ

α
(cφ
w
γ
). (10.60)
Expression (10.59) is the classical diffusion equation wherein the flux of
matter is given by J
γ
= ρc(1 − c)φ
w
γ
. Relation (10.60) represents the
274 10 Hamiltonian Formalisms
most original contribution of GENERIC formalism to this problem, as
it is a time evolution equation for the relative velocity w, which has no
counterpart in the classical theory of irreversible processes. More about its
physical content will be discussed below.
4. Dissipation
As viscous contributions are neglected, the dissipation potential is simply
given by
ψ =
1
2
φ
w
α
Λφ
w
α
, (10.61)
where Λ is a phenomenological coefficient which is positive to guarantee
the positiveness of the dissipation potential. The complete set of evolution
equations is still given by (10.58)–(10.60) at the condition to complement
the evolution equation (10.60) for w
α
by the dissipative term −Λφ
w
α
.
It is interesting to observe that by inserting this dissipative term in (10.60),
one obtains an equation of the Cattaneo type. By omitting in (10.60) the
non-linear contributions in the fluxes and making use of the result φ
w
α
=
w
α
/ρc(1 −c), one arrives at

t
w
α
+ [Λ/ρc(1 −c)]w
α
= −ρc(1 −c)∂
α

c
/ρ), (10.62)
which is clearly of the form
τ∂
t
w
α
+w
α
= −D


α

c
/ρ), (10.63)
where τ = 2ρc(1 −c)Λ
−1
corresponds to the relaxation time of the diffusive
momentum and D

= [ρc(1 − c)]
2
Λ
−1
to a diffusion-like coefficient. Expres-
sion (10.63) is an example of non-Fickian law. Non-Fickian diffusion occurs for
instance when a solvent diffuses into a medium characterized by a timescale
which is the same or larger than the timescale on which the penetration
takes place. This situation arises when solvents penetrate into glassy poly-
mers. Another example in which diffusion exhibits non-classical features is the
mixture of
3
He–
4
He isotopes below the lambda point (Lhuillier et al. 2003;
Lebon et al. 2003).
If we assume that w
α
evolves much faster than the other state variables
(∂
t
w
α
=0), then (10.63) reduces to the standard description of Fick’s dif-
fusion. Indeed, in virtue of the definition of the diffusion flux J
α
= ρc(1 −
c)φ
w
α
≡ w
α
, (10.63) becomes simply
J
α
= −D


α

c
/ρ), (10.64)
which is the classical Fick’s law.
10.3 Final Comments
In this section, GENERIC has been presented as a general formalism to de-
scribe systems driven out of equilibrium. The main feature of this theory is
10.3 Final Comments 275
that it uses two generating potentials: the total energy for reversible dynamics
and entropy (or more generally the dissipation potential) for irreversible dy-
namics. The principal objective is to derive the relevant evolution equations
of the process under study. The main result of the study is that it offers a
structure which is applicable whatever the level of description, either macro-
scopic, mesoscopic, or microscopic. The basic ideas underlying GENERIC
have been discussed thanks to two simple examples: classical hydrodynamics
and diffusion of two perfectly miscible fluids.
It is instructive to recapitulate how we obtained the time evolution equa-
tions. First, one has to define the space of the state variables x. This choice
depends on our goals and the nature of the system under investigation; it
is made on the basis of some intuition guided by our insight of the physics
of the problem. In classical hydrodynamics, the solution is unambiguous:
the state variables are the classical hydrodynamic fields, i.e. mass, momen-
tum, and energy or temperature. However, this classical approach fails when
dealing with complex fluids, like non-Newtonian or polymer solutions. Due
to the active role of the microscopic structure of the complex fluids in the
macroscopic time evolution, the choice of the state variables is not universal.
For example, to describe complex fluids, the hydrodynamic variables will be
complemented by structural variables, sometimes called internal variables as
in Chap. 8. This is in opposition with the use of functionals as in rational
thermodynamics. We emphasize that the choice of state variables represents
always a first step to make and that this is always an assumption. It should
also be stressed that GENERIC does not offer any help in the selection of
the basic variables.
The next step is to formulate the corresponding set of evolution equations.
In classical hydrodynamics, they are the usual balance equations of mass,
momentum, and energy complemented by appropriate constitutive equations
allowing us to “close” the set of balance equations. In GENERIC, one makes
a clear distinction between the reversible and the irreversible contributions
to the kinematics of the field variables. The non-dissipative part is obtained
explicitly as a straightforward consequence of the dynamics expressed by
the Poisson brackets or equivalently by the skew-symmetric operator
ˆ
L. The
reversible part of the evolution equation is completely specified by the kine-
matics, which means that explicit relations are obtained for the evolution
equations including expressions for the scalar thermostatic pressure and even-
tually other extra reversible contributions like reversible stress tensors. After
the evolution equations have been derived, we have no freedom to use other
physical requirements to cancel, add or modify the governing evolution re-
lations. The only thing that we can change is the Poisson bracket, indeed
we cannot exclude the possibility that other Poisson brackets, different from
these used in the preceding sections, provide a better physical description of
the physical process.
How do we find the Poisson brackets? A first type of arguments consists
in proceeding by trials and errors and, guided by our physical intuition, by
276 10 Hamiltonian Formalisms
comparing the results as arising from the Poisson kinematics with results de-
rived from other approaches. A second more sophisticated method consists in
using the Lie group of transformations as addressed in
¨
Ottinger’s book (2005)
to which we refer for more details. The dissipative part of the evolution equa-
tions, as specified by bracket (10.20) is derived after an explicit expression for
the dissipation potential ψ has been determined, the latter being compelled
to obey the properties to be zero at equilibrium and to be a convex function
of the variables outside equilibrium. To establish a valuable expression for
the dissipation potential is, generally speaking, not an easy task as it should
involve material information related to non-equilibrium parameters as vis-
cosity, diffusion coefficient, thermal conductivity, hydrodynamic interaction
tensors, etc.
In short, the formulation of the evolution equations of GENERIC requires
the knowledge of the following entities: the Poisson bracket, the energy E
(or an equivalent thermodynamic potential as Helmholtz free energy), the
dissipation potential ψ, and entropy S. The three functions E, S and ψ
contain the individual features of the medium under study while to be sure
that we are in presence of a true Poisson bracket, we have to check the
Jacobi’s identity ¦A, ¦B, C¦¦ + ¦B, ¦C, A¦ + ¦C, ¦A, B¦¦; this represents a
heavy and tedious task but is of importance because it expresses that the
reversible dynamics has a Hamiltonian structure. In recent publications about
GENERIC, there is a tendency in favour of the use of the following building
blocks: the two generators E (energy) and S (entropy) and the two operators
ˆ
L and
ˆ
M.
As shown in Sect. 10.3, the splitting of the total pressure tensor as

αβ
+ P
αβ
into a reversible part pδ
αβ
and an extra viscous tensor P
αβ
arises as a result of the GENERIC approach. This is not so in the con-
text of other formalisms as classical irreversible thermodynamics, extended
irreversible thermodynamics or rational thermodynamics, where this split-
ting is imposed a priori. Moreover, in most non-equilibrium thermodynamic
theories, it is the entropy production that is put into the focus; the part that
does not produce entropy, the so-called gyroscopic forces of the form ω x
where ω is the angular velocity, or the convected parts of the generalized
time derivatives remain largely undetermined. It is a success of GENERIC to
incorporate naturally such gyroscopic effects thanks to the Poisson structure
of the evolution equations and to propose explicit relations for these non-
dissipative contributions. In that respect, it is worth noting that Jacobi’s
identity imposes that the convected derivatives fulfil the principle of material
frame-indifference; this exhibits the strong link between Jacobi’s identity and
frame-indifference and sheds a new lighting on this controversial principle (see
Sect. 9.5).
Note also that in GENERIC, even the well-known and well-tested balance
equations of mass, momentum, and energy are re-derived while such relations
are usually taken for granted in other thermodynamic theories. Moreover,
the governing equations in GENERIC are generally non-linear and this is the
10.3 Final Comments 277
reason why they find a natural domain of application in rheology. Another
difference between GENERIC and some other thermodynamic approaches
is that in the latter, the Onsager reciprocal relations are only applied in
the linear regime; in GENERIC, this property is taken for granted even in
the non-linear domain as attested by the requirement that the symmetry of
operator
ˆ
M is to be universally satisfied.
It should also be realized that in GENERIC, it is generally understood
that the system under consideration is isolated from its environment as it is
assumed that the total energy is conserved dE/dt = 0 and the total entropy
is increasing with time dS/dt > 0. This means that the theory is not well
adapted for describing open and/or externally forced systems as for instance
B´enard’s convection in a fluid submitted to an external temperature gradient.
In the case of non-isolated systems, one should, as in classical thermodynam-
ics, introduce some kind of hypothetical reservoirs but such an operation
may be delicate and obscure the formalism (Grmela 2001). In a recent paper,
¨
Ottinger (2006) suggests to split the Poisson and dissipative brackets into
bulk and boundary contributions.
Moreover, in presence of coupled heat and mass transport, the procedure
dictated by GENERIC is rather intricate and requires the introduction of
unusual state variables like some particular combinations of the entropy flux
and the mass flux (Grmela et al. 2003). Finally, similarly to other approaches,
GENERIC formalism remains silent about the physical meaning of some basic
quantities such as non-equilibrium temperature.
GENERIC has been the subject of several applications in statistical me-
chanics, molecular dynamics, quantum mechanics and macroscopic physics,
mainly in rheology (
¨
Ottinger 2005) where one of the most important prob-
lems is to derive constitutive equations consistent with the dynamics of the
basic variables. A number of new results produced by GENERIC should also
be mentioned, for instance the formulation of generalized reptation models
(
¨
Ottinger 1999a), of equations in relativistic hydrodynamics and cosmology
(Ilg and
¨
Ottinger 1999;
¨
Ottinger 1999b), new models for heat transfer in
nano-systems (Grmela et al. 2005). GENERIC may also be considered as
an extension of the bracket formulation as proposed by Beris and Edwards
(1994) and Beris (2003).
The essential merit of GENERIC is its unifying universality as it is applica-
ble whatever the level of description of the material system, either macro-
scopic, mesoscopic, or microscopic. It should also be realized that GENERIC,
as well as extended thermodynamics, internal variables theories or rational
thermodynamics, are not exclusive of each other: they complement harmo-
niously and offer, in most circumstances, results that are in agreement with
each other. Finally, yet importantly, it is expected that a better comprehen-
sion of all these theories will provide a step forward the formulation of a
desirable unified theory of irreversible processes.
278 10 Hamiltonian Formalisms
10.4 Problems
10.1. Poisson bracket. Show that the scalar product (∂A/∂x,
ˆ
L∂B/∂x) with
ˆ
L =
_
0 1
−1 1
_
is equivalent to the Poisson bracket ¦A, B¦.
10.2. Jacobi’s identity. Verify that by taking
ˆ
L =
_
0 1
−1 0
_
, the bracket (10.4)
satisfies Jacobi’s identity (10.8).
10.3. Degeneracy conditions. Prove that the results dE/dt = 0, dS/dt >0
may be considered as a direct consequence of the degeneracy conditions
ˆ
LδS/δx = 0 and
ˆ
M δE/δx = 0.
10.4. Degeneracy conditions. Check that the operators
ˆ
L and
ˆ
M given by
(10.46) verify the degeneracy properties
ˆ
LδS/δx = 0,
ˆ
M δE/δx = 0.
10.5. Isothermal fluid. Show that in the case of a one-component isothermal
fluid characterized by the variables x = (ρ, u
α
), the functional derivatives
δE/δx and δS/δx are given by
δE
δx
=
_
−u
2
/2ρ +∂ε/∂ρ
u
α

_
,
δS
δx
=
_
−µ/T
0
_
,
wherein µ designates the chemical potential and T the uniform temperature.
10.6. Non-isothermal fluid. Derive the GENERIC building blocks E, S, L and
M for a one-component fluid in presence of temperature gradients (
¨
Ottinger
and Grmela 1997).
10.7. Matter diffusion. Establish the relevant equations of matter diffusion
in binary mixtures in the simplified case of non-viscous fluids and uniform
temperature.
10.8. Non-Fickian diffusion. Repeat the analysis of Sect. 10.2.2 when it is
assumed that the barycentric velocity u does not vanish.
10.9. Thermodiffusion. Consider the more complex problem of coupling of
heat and mass transport (thermodiffusion) in a two-component system. De-
termine the evolution equations for the total mass density, the mass fraction,
the momentum, and the energy in the framework of the above Hamiltonian
formalism. Hint: See Grmela et al. (2003).
10.10. Extended irreversible thermodynamics. Show that the evolution equa-
tions governing the behaviour of the dissipative fluxes of extended irreversible
thermodynamics possess a Hamiltonian structure (Grmela and Lebon 1990).
Chapter 11
Mesoscopic Thermodynamic Descriptions
Thermodynamics and Fluctuations
The previous chapters were essentially concerned with the average behaviour
of thermodynamic variables without reference to their fluctuations. The lat-
ter are not only a subject of intense study in statistical mechanics, but also
play a significant role in non-equilibrium thermodynamics, like near critical
points, non-equilibrium instabilities and, more generally, in processes taking
part at small time and space scales. They played an important role in the
original derivation of Onsager–Casimir’s reciprocal relations, and in modern
formulations of the fluctuation–dissipation relations expressing the transport
coefficients in terms of the correlation functions of the fluctuations of the
fluxes. Furthermore, when the fluctuations become important, they must be
included amongst the set of independent variables. The status of fluctua-
tions being intermediate between the macroscopic and the microscopic, it is
justified to refer to them as mesoscopic variables.
In the first part of this chapter is presented an introductory overview of the
theory of fluctuations. After recalling Einstein’s theory of fluctuations around
equilibrium, we derive Onsager–Casimir’s reciprocal relation, and discuss the
fluctuation–dissipation theorem, which is important in non-equilibrium sta-
tistical mechanics. Afterwards, we analyse two formalisms which have in-
fluenced the recent developments of irreversible thermodynamics, namely
Keizer’s theory, where a non-equilibrium entropy is defined in terms of the
second moments of fluctuations, and the so-called mesoscopic thermodynam-
ics, in which the distribution function of the fluctuations is itself the central
variable.
11.1 Einstein’s Formula: Second Moments
of Equilibrium Fluctuations
In 1902, Einstein proposed a relation between the probability of the fluctu-
ation of a thermodynamic variable with respect to its equilibrium value and
the entropy change in isolated systems. This was obtained by inverting the
279
280 11 Mesoscopic Thermodynamic Descriptions
well-known Boltzmann–Planck’s formula
S = k
B
ln W, (11.1)
where S is the total entropy of the system, W is the probability of the
macrostate, and k
B
is the Boltzmann constant. With Einstein, let us write
W ≈ exp(∆S/k
B
), (11.2)
where ∆S is the change of entropy associated to the fluctuations, namely
∆S = S −S
eq
. (11.3)
As an example, consider the fluctuations of energy in an isolated system
composed of two subsystems in thermal contact. In equilibrium, both of them
have the same temperature, and the total internal energy is distributed in
such a way that the total entropy is maximum. However, despite the total
internal energy remains constant, both subsystems may exhibit fluctuations
in their respective internal energies.
If the fluctuations are small, ∆S may be expanded around equilibrium, so
that
∆S ≈ (δS)
eq
+
1
2

2
S)
eq
, (11.4)
δS and δ
2
S being the first and the second differentials of the entropy. For
an isolated system, one has (δS)
eq
= 0 and (δ
2
S)
eq
≤ 0, because the total
entropy is maximum at equilibrium. Introduction of these results into (11.2)
yields the well-known Einstein’s formula for the probability of fluctuations,
W ≈ exp
_
1
2
δ
2
S
k
B
_
. (11.5)
Note that this relation is restricted to second-order developments and is only
applicable to isolated systems.
To extend (11.5) to non-isolated systems, Einstein suggested to express
the probability of fluctuations as
W ≈ exp
_

∆/
k
B
T
0
_
, (11.6)
where /, the availability, is a new quantity defined as
/ = U −T
0
S +p
0
V, (11.7)
in which T
0
and p
0
are the temperature and pressure of the environment.
Thus, / is not an intrinsic property of the system alone but of the ensemble
formed by it and its surroundings. It can be shown that ∆/ measures the
maximum amount of useful work which can be extracted from the system
during exchanges with the outside world. If the system and the surroundings
11.1 Einstein’s Formula: Second Moments of Equilibrium Fluctuations 281
are in thermal and mechanical equilibrium, T = T
0
and p = p
0
, the change
in / is given by
∆/ = ∆U −T∆S +p∆V = ∆G, (11.8)
where G is the Gibbs’ function, so that (11.6) takes the form
W ≈ exp(−∆G/k
B
T). (11.9)
Similarly, for a system at fixed V and T, ∆/ = ∆F, with F the Helmholtz
free energy, and the probability of fluctuations becomes
W ≈ exp(−∆F/k
B
T). (11.10)
For small fluctuations, ∆/ can be expanded as
∆/ ≈ δ/
eq
+
1
2

2
/)
eq
+ ,
where the first-order term is zero because at equilibrium G is minimum at
fixed T and p (or, respectively, F at fixed V and T). As a consequence,
W ≈ exp
_

1
2
δ
2
G
k
B
T
_
, (11.11a)
W ≈ exp
_

1
2
δ
2
F
k
B
T
_
. (11.11b)
These equations can still be expressed in terms of δ
2
S because δ
2
F at constant
T and V (or δ
2
G at fixed p and T) is proportional to δ
2
S. Indeed, remember
that F = U−TS; when one studies the fluctuations of F at constant T and V ,
U is the independent variable and S is a function of U so that δ
2
F = −Tδ
2
S.
From (11.11a), one recovers (11.5), but with δ
2
S computed at constant T
and V . Similarly, it is easy to see that δ
2
G = −Tδ
2
S when combined with
(11.11b), again leads to (11.5) but with δ
2
S computed at constant T and p.
These observations indicate that (11.5) is applicable even outside isolated
systems, at the condition to specify explicitly which variables are kept fixed.
In an isolated system, information about the fluctuations around equi-
librium states implies the knowledge of δ
2
S in terms of δU and δV . The
second variation of S is directly obtained from the Gibbs’ equation δS =
T
−1
δU +pT
−1
δV which, introduced in (11.5) yields
W ∝ exp
_

1
2k
B
_
(T
−1
)
U
(δU)
2
+ 2(T
−1
)
V
δUδV + (T
−1
p)
V
(δV )
2
¸
_
,
(11.12)
where subscripts U and V stand for partial differentiation with respect to
these variables.
Recall the mathematical result that for a Gaussian distribution function
of the form
W ≈ exp(−
1
2
E
ij
δx
i
δx
j
), (11.13)
282 11 Mesoscopic Thermodynamic Descriptions
the second moments of the fluctuations are given by
¸δx
i
δx
j
) = (E
−1
)
ij
, (11.14)
where δx
i
designates the fluctuation of x
i
and the brackets ¸. . .) denote the
average over the probability distribution (11.13). The second moments of the
fluctuations of U and V may be obtained from (11.12) and (11.14) by taking
into account that the matrix corresponding to the second derivatives of the
entropy with respect to these variables is
E
UV
= −
1
k
B
_
(T
−1
)
U
(T
−1
)
V
(T
−1
p)
U
(T
−1
p)
V
_
. (11.15)
By inverting the matrix and introducing the result in (11.14), it is left as an
exercise to prove that (see Problem 11.1)
¸δUδU) = k
B
C
p
T
2
−2k
B
T
2
pV α +k
B
Tp
2
V κ
T
,
¸δUδV ) = k
B
T
2
V α −k
B
TpV κ
T
,
¸δV δV ) = k
B
TV κ
T
,
(11.16)
where C
p
is the heat capacity at constant pressure; α is the coefficient of
thermal expansion, and κ
T
is the isothermal compressibility. Near critical
points, when some of the coefficients in (11.16) diverge, the second moments
of the fluctuations become very large and they have a clear macroscopic
influence on the behaviour of the system. The analysis of this behaviour has
been a main topic of research in statistical mechanics from the 1960s to 1980s
(e.g. Callen 1985; Landau and Lifshitz 1980; Reichl 1998).
There is a natural distinction in the fluctuation theory between statics
and dynamics. The former is related to properties occurring at equal times,
as discussed in the present section, the latter is concerned with their evo-
lution during finite time intervals. This aspect is discussed in the next sec-
tions in relation with the derivation of Onsager–Casimir’s relations and the
dissipation–fluctuation theorem.
11.2 Derivation of the Onsager–Casimir’s Reciprocal
Relations
The analysis of the evolution of fluctuations in the course of time led On-
sager to establish his famous reciprocal relations, which were introduced as
a phenomenological postulate in Chap. 2, and whose microscopic derivation
is outlined here.
Assume that the entropy of an isolated system is a function of several
variables A
1
, . . . , A
n
and denote by α
β
the fluctuations from their average
11.2 Derivation of the Onsager–Casimir’s Reciprocal Relations 283
equilibrium value ¸A
β
), i.e. α
β
= A
β
− ¸A
β
). For small fluctuations around
equilibrium, the entropy may be written as
S = S
eq

1
2
G
βγ
α
β
α
γ
, (11.17)
with
G
βγ
=
_

2
S
∂A
β
∂A
γ
_
eq
, (G
βγ
= G
γβ
). (11.18)
Summation with respect to repeated subindices is understood. It follows from
Einstein’s relation (11.5) that the second moments of the fluctuations are
given by
¸α
β
α
γ
) = k
B
(G
−1
)
βγ
. (11.19)
In his original demonstration, Onsager assumed that the decay of spon-
taneous fluctuations of the system obeys the same phenomenological laws
that describe the evolution of the perturbations produced by external causes.
Furthermore, he supposed that these laws are linear, so that the evolution of
α
β
is governed by

β
dt
= −M
βγ
α
γ
(11.20)
with M
βγ
a matrix independent of α
γ
.
Moreover, Onsager defines the thermodynamic fluxes J
β
as the time deriv-
atives of the fluctuations α
β
and the thermodynamic forces X
β
as the deriv-
atives of the entropy with respect to α
β
, i.e.
J
β
=

β
dt
, (11.21a)
X
β
=
_
∂∆S
∂α
β
_
= −G
βγ
α
γ
. (11.21b)
With these definitions of J
β
and X
β
, the time derivative of the entropy is
bilinear in the fluxes and the forces
dS
dt
= −G
βγ
α
γ

β
dt
= J
β
X
β
, (11.22)
while the evolution equations (11.20) can be cast in the form of linear relations
between fluxes and forces:
J
β
= L
βγ
X
γ
. (11.23)
Indeed, it suffices to introduce the definitions (11.21) into (11.23) to obtain

β
dt
= −L
βγ
G
γη
α
η
. (11.24)
Comparison with (11.20) leads to the identification
L = M G
−1
. (11.25)
284 11 Mesoscopic Thermodynamic Descriptions
In a steady state, which by definition is invariant with respect to a translation
in time, one has for any couple of variables α
β
and α
γ
¸α
β
(0)α
γ
(t)) = ¸α
β
(0 +τ)α
γ
(t +τ)). (11.26)
Since this equality is valid for any values of t and τ; it is in particular true
for τ = −t, from which
¸α
β
(0)α
γ
(t)) = ¸α
β
(−t)α
γ
(0)). (11.27)
Assume that the variable A
β
has a well-defined time-reversal parity under
the transformations t → −t, by definition, even variables will transform as
α
β
(−t) = α
β
(t), while odd variables will behave as α
β
(t) = −α
β
(−t). Mass
and energy are examples of even variables, and momentum and heat flux are
odd variables. This allows us to write (11.27) in whole generality as
¸α
β
(0)α
γ
(t)) = ε
β
ε
γ
¸α
β
(t)α
γ
(0)), (11.28)
where ε
β
= +1 or −1 according the variable is even or odd. Note that (11.28)
implies that equal-time correlations between even and odd variables vanish
in equilibrium. Relation (11.28) expresses the microscopic reversibility of the
system and is of fundamental importance in deriving the Onsager–Casimir’s
relations.
Moreover, direct integration of the evolution equation (11.20) yields
α
β
(t) = exp[−Mt]
βγ
α
γ
(0). (11.29)
For small values of t, one may expand (11.29) in Taylor’s series, by keeping
only first-order terms in t, and (11.28) takes then the form
¸α
β
(0)(δ
γη
−M
γη
t)α
η
(0)) = ε
β
ε
γ
¸(δ
βη
−M
βη
t)α
η
(0)α
γ
(0)), (11.30)
from which
M
γη
¸α
β
(0)α
η
(0)) = ε
β
ε
γ
M
βη
¸α
η
(0)α
γ
(0)). (11.31)
since δ
γη
= 1 if γ = η, δ
γη
= 0 otherwise. Expressing the second moments in
terms of the matrix G of the second derivatives of the entropy, (11.31) can
be written as
M
γη
(G
−1
)
βη
= ε
β
ε
γ
M
βη
(G
−1
)
ηγ
. (11.32)
From relation (11.25) between L and M, and the property that G is a sym-
metric matrix by construction, it follows that
L
γβ
= ε
β
ε
γ
L
βγ
, (11.33)
11.3 Fluctuation–Dissipation Theorem 285
which are the well-known Onsager–Casimir’s reciprocal relations. Historically,
Onsager derived the reciprocity relations in 1931, but only in the case of even
variables. The extension to odd variables was achieved in 1945 by Casimir.
The above analysis exhibits clearly the main assumptions underlying the
derivation of the Onsager–Casimir’s relations, namely linear evolution equa-
tions and the hypothesis that the behaviour of fluctuations is described on the
average by relations mimicking the macroscopic transport laws. In Chap. 2
these relations were presented as a macroscopic postulate, whereas here it
follows from dynamical equations and statistical considerations.
In presence of a magnetic field H, the operation of time reversal will only
change the internal magnetic field arising from the motion of the particles,
which is automatically reversed under time reversal. However, because of
the Lorentz force, which is proportional to the vector product v H of the
velocities and the magnetic induction, the external magnetic field H must
be inverted in order that Lorentz’s force remains unchanged; this guarantees
that the reversal motion follows the same path as the direct one. The same
comment applies when the system rotates as a whole with angular velocity
ω; the sign of the angular velocity must be changed because the Coriolis
force is proportional to the vector product ω v of the velocities of the
particles and the angular velocity of the system. Therefore, in presence of
an external magnetic induction H and angular rotation ω, the Onsager–
Casimir’s relations read as
L
γβ
(H, ω) = ε
β
ε
γ
L
βγ
(−H, −ω). (11.34)
The importance and the limitations of the Onsager–Casimir’s relations have
been underlined in Chap. 2. Here, the role of the time-reversal parity of the
several variables becomes clear, as well as the reason why H and ω must be
inverted to keep the validity of the reciprocal relations.
11.3 Fluctuation–Dissipation Theorem
In his paper on Brownian motion published in 1905, Einstein established that
molecular collisions between the fluid and the Brownian particles are the
source of friction or dissipation and moreover responsible for the occurrence
of fluctuations in the position of the particles. Einstein’s idea was further
explored by many authors, and was the basis for the derivation of the so-called
fluctuation–dissipation theorem, expressing that fluctuations and dissipative
coefficients are related to each other. Langevin’s stochastic analysis of the
motion of a Brownian particle, and Nyquist study of the fluctuations of the
electric current in a conductor contributed further to a better understanding
of the relation between fluctuations and dissipation. Since the 1950s, these
analyses were set on a more general basis by Callen and Welton (1951),
286 11 Mesoscopic Thermodynamic Descriptions
Callen and Greene (1952), and Tisza and Manning (1957). It is not here our
purpose to enter into the details of their derivations, but rather to underline
the general ideas, because of their importance in statistical thermodynamics.
One of the main objectives of non-equilibrium statistical mechanics is to
provide, at the microscopic level, an expression for the memory function
relating a flux J(t), at time t, to its conjugate force X(t

), at previous times
t

≤ t:
J(t) =
_
0
−∞
K(t −t

) X(t

)dt

; (11.35)
K(t −t

) is the so-called memory function or memory kernel. As an example,
consider the Maxwell–Cattaneo evolution equation written formally as,
τ
∂J
∂t
= −(J −L X), (11.36)
with L the corresponding transport coefficient. Integration of (11.36) yields,
for a zero initial flux at time t

= −∞,
J(t) =
_
0
−∞
L
τ
exp [−(t −t

)/τ] X(t

)dt

. (11.37)
By comparing (11.37) and (11.35), one may identify the memory function as
K(t −t

) =
L
τ
exp
_

t −t

τ
_
, (11.38)
indicating that in the linearized version of extended irreversible thermo-
dynamics (EIT), the memory function decays exponentially. More general
versions of transport theory can be found in modern statistical mechanics,
based on N-particle distribution functions instead of the more restrictive
one-particle distribution function, and on the Liouville equation for the de-
scription of its evolution.
The most significant feature of the fluctuation–dissipation theorem is that
the memory kernel is related to the correlation function of the fluctuations
of the corresponding fluxes through (McQuarrie 1976; Reichl 1998)
K(t −t

) =
1
k
B
T
¸δJ(t)δJ(t

)), (11.39)
angular brackets stand for the average over an equilibrium distribution
function.
As particular examples of (11.39), the memory functions generalizing the
thermal conductivity and shear viscosity are written as
11.3 Fluctuation–Dissipation Theorem 287
λ
ij
(t −t

) =
V
k
B
T
2
¸δq
i
(t)δq
j
(t

)), (11.40)
η
ijkl
(t −t

) =
V
k
B
T
¸δP
v
ij
(t)δP
v
kl
(t

)). (11.41)
The knowledge of the quantities inside the brackets of (11.40) and (11.41)
in microscopic terms opens the way to the determination of the memory
functions in the most general situations, from dilute ideal gases to dense
fluids.
In terms of the memory functions (11.40) and (11.41), the Fourier’s and
Newton–Stokes’ equations are generalized in the form
q
i
(t) = −
_
0
−∞
λ
ij
(t −t

)
∂T(t

)
∂x
j
dt

, (11.42a)
P
v
ij
(t) = −
_
0
−∞
η
ijkl
(t −t

)V
kl
(t

)dt

. (11.42b)
When the relaxation times of the fluctuations of the fluxes are very short, in
such a way that the memory functions are negligible except for very small
values of t −t

, the bounds of the integrals may be extended to infinity and
the thermodynamic forces may be taken out of the integral with the value
corresponding to t = t

. Doing so, one recovers from (11.40) and (11.41), the
Green–Kubo’s formula relating the dissipative coefficients of heat conductiv-
ity and shear viscosity to the fluctuations of the heat flux and the momentum
flux, respectively:
λ
ij
= (k
B
T
2
)
−1
_

0
¸δq
i
(0)δq
j
(t))dt, (11.43a)
η
ijkl
= (k
B
T)
−1
_

0
¸δP
v
ij
(0)δP
v
kl
(t))dt. (11.43b)
The Green–Kubo’s relations are useful for calculating the dissipative trans-
port coefficients from first principles, independently of the density of the sys-
tem, in contrast with the classical calculations starting from the Boltzmann
equation, whose validity is limited to a low-density range (Reichl 1998).
Despite their elegance and generality, the simplicity of the above results
is misleading because of the complexity of the microscopic Liouville oper-
ator describing the evolution of the fluctuations of the fluxes. This is why
some approximations must be introduced, usually in the form of mathemat-
ical models for the memory functions, as a prerequisite for solving practical
problems. Therefore, formalisms such as EIT may be of valuable assistance
to model the evolution of the fluctuations of the fluxes.
288 11 Mesoscopic Thermodynamic Descriptions
11.4 Keizer’s Theory: Fluctuations in Non-Equilibrium
Steady States
11.4.1 Dynamics of Fluctuations
Keizer (1978, 1983, 1987) proposed to regard the fluctuations as the founda-
tions for a definition of entropy outside equilibrium. More recently (Reguera
2000, Rubi 2004), attempts have been made to upgrade the fluctuations to the
status of independent variables besides their usual average values. This topic
will be analysed in the next section.
Keizer combines the molecular picture of Boltzmann and the stochastic
picture of Onsager, and lays the foundations of a statistical thermodynam-
ics of equilibrium and non-equilibrium steady states. Steady states are, like
equilibrium states, time-independent but they involve constant inputs and
outputs per unit time, i.e. they require more parameters, like the fluxes, to
specify the state.
Consider a large molecular system undergoing a series of elementary
processes characterized by a set of extensive variables x(= U(energy),
V, N . . . N (volume), N
1
, . . . , N
n
, with N
i
the number of particles of com-
ponent i). According to the Boltzmann–Planck’s postulate, the probability
of a fluctuation δx around local equilibrium is given by
W(δx) ≈ exp(∆S/k
B
) = ¦(2π)
−(n+2)
det[(−S
eq
/k
B
)]¦
1/2
exp(δ
2
S
eq
/2k
B
),
(11.44)
with
(S
eq
)
ij
= (∂
2
S/∂x
i
∂x
j
)
eq
, δ
2
S
eq
= (∂
2
S/∂x
i
∂x
j
)
eq
δx
i
δx
j
, (11.45)
after performing a Taylor expansion around equilibrium and using the max-
imum entropy principle. Like in Sect. 11.1, δx stands for the fluctuations of
the variables with respect to their equilibrium values (namely δx(r, t) ≡
x(r, t) −x
eq
≡ (δU, δV, δN
1
, . . . , δN
n
). From now on, we shall omit for sim-
plicity to write the time and position dependence of the variables.
A main feature of Keizer’s theory is that the elementary molecular
processes can be described by “canonical” kinetic evolution equations of the
form
dx
i
dt
=

k
ω
ki
(V
+
k
−V

k
) +J
i
(x, t) ≡ R
i
(x, t). (11.46)
The first term in the right-hand side represents the dissipative part of the
evolution and J
i
the non-dissipative part, R
i
is a compact notation to express
the sum of both terms. The quantity ω
ki
stands for ω
ki
= x
+
ki
− x

ki
, with
11.4 Keizer’s Theory: Fluctuations in Non-Equilibrium Steady States 289
the + or − upper indices denoting the forward or backward variation of the
values of the extensive variable i in the elementary process k, finally V
±
k
is
given by
V
±
k
= Ω
±
k
exp



1
k
B

j
x
±
kj
_
∂S
∂x
j
_


, (11.47)
with Ω
±
k
the intrinsic forward and backward rates of the elementary k process.
It was shown by Keizer that the Fourier’s, Fick’s, and Newton–Stokes’ trans-
port laws as well as the chemical mass action law or the Boltzmann collision
operator may be written in the canonical form (11.46).
It follows from (11.46) that the dynamics of fluctuations around a steady
state may be described by
dδx
i
dt
=
_
∂R
i
∂x
j
_
δx
j
+f
i
≡ H
ij
(x
ss
)δx
j
+f
i
, (11.48)
where superscript ss stands for steady state, and f
i
for a stochastic noise. By
definition, one has ¸f
i
) = 0 and
¸f
i
(t)f
j
(t

)) =

j
ω
ki
_
V
+
k
(x
ss
) +V

k
(x
ss
)
¸
ω
kj
δ(t −t

) ≡ γ
ij
(x
ss
)δ(t −t

).
(11.49)
It follows from these results that the probability of fluctuations in a non-
equilibrium steady state may be written in the Gaussian form
W(δx) = [(2π)
−(n+2)
det σ
−1
] exp
_

1
2
δx
T
σ
−1
δx
_
, (11.50)
with σ the covariance matrix,
σ ≡ ¸δxδx)
ss
, (11.51)
given by the generalized fluctuation–dissipation expression
H(x
ss
) σ + σ H
T
(x
ss
) = −γ(x
ss
). (11.52)
Since the matrices H and γ are known from (11.48) and (11.49), the non-
equilibrium covariance matrix (11.51) may directly be obtained from (11.52).
11.4.2 A Non-Equilibrium Entropy
An important pillar of Keizer’s theory is the definition of a non-equilibrium
entropy function S
K
by interpreting (11.50) in an analogous way to (11.44),
namely as

2
S
K
∂x
i
∂x
j
≡ −k
B

−1
)
ij
. (11.53)
290 11 Mesoscopic Thermodynamic Descriptions
This is a differential equation for S
K
, which may be solved provided that
∂σ
ij
/∂x
k
= ∂σ
ik
/∂x
j
, a condition that was explicitly studied by Keizer. At
equilibrium, σ reduces to the equilibrium covariance matrix and S
K
becomes
identical to the equilibrium entropy. The physics underlying the above defi-
nition is that the entropy S
K
is related to the fluctuations in the extensive
variables in a way analogous to Boltzmann–Planck’s formula (11.1).
When expressed in terms of S
K
, the statistical distribution (11.53) is of
the form W ≈ exp(δ
2
S
K
/2k
B
) with
δ
2
S
K
=

ij
(∂
2
S
K
/∂x
i
∂x
j
)
ss
(x
i
−x
ss
i
)(x
i
−x
ss
i
). (11.54)
Since stability requirements impose that σ must be a positive definite matrix,
it follows from (11.53) that in stable steady states δ
2
S
K
≤ 0. Moreover, it was
shown by Keizer that the second variation of S
K
is a Lyapounov function,
from which
δ
2
S
K
≤ 0,
d
dt
δ
2
S
K
≥ 0. (11.55)
Since non-equilibrium states can only be sustained by the presence of fluxes
(of energy, mass, etc.) between the system and reservoirs (of energy, mass,
etc.), it is assumed by Keizer that S
K
will depend not only on the intensive
variables x but also on the fluxes f and, in addition, some intensive variables
Γ characterizing the reservoirs, so that
S
K
= S
K
(x; f , Γ). (11.56)
In analogy with classical thermodynamics, Keizer defines intensive vari-
ables conjugate to the extensive variables x
i
by
φ
i
=
∂S
K
∂x
i
. (11.57)
In particular, the intensive variables conjugate to the energy, volume and
number of moles are related to generalized temperature T, pressure p, and
chemical potentials µ
i
, in such a way that the generalized Gibbs’ equation
for S
K
will read as
dS
K
= T
−1
dU +pT
−1
dV −

i
µ
i
T
−1
dN
i
+

j
∂S
K
∂f
j
df
j
+

l
∂S
K
∂Γ
l

l
.
(11.58)
Note that (11.58) contains a contribution in the fluxes, which is analogous
to EIT – the latter lacking the last term, related to the properties of the
reservoirs.
Expanding S
K
(x; f , Γ) around equilibrium results in
S
K
(x; f , Γ) = S
eq
(x) −

i
f
i
v
i
(x; f , Γ), (11.59)
11.4 Keizer’s Theory: Fluctuations in Non-Equilibrium Steady States 291
where ν
i
is the “force” conjugated to f
i
. As a consequence, any intensive
variable, like for instance the temperature T conjugated to the internal energy
U through 1/T = ∂S
K
/∂U, will be given by
1
T
=
1
T
eq

i
f
i
∂v
i
∂U
. (11.60)
This result shares some features with EIT, as it exhibits the property that
the temperature is not equal to the (local) equilibrium temperature T
eq
, but
contains additional terms depending on the fluxes (see Box 7.3).
An alternative expression of the generalized entropy, more explicit in the
fluctuations and based on (11.53), is
S
K
(x; f , Γ) = S
eq
(x) −
1
2
k
B

ij
x
i
x
j

−1
−σ
−1
eq
)
ij
, (11.61)
where x
0
= U corresponds to the internal energy. It is directly inferred from
(11.61) that the temperature can be cast in the form
1
T
=
1
T
eq
−k
B

i
x
i

−1
i0
−σ
−1
i0,eq
). (11.62)
This result is important as it points out that non-equilibrium temperature
(the same reasoning remains valid for the pressure or the chemical potential)
is not identical to the local equilibrium one but is related to it through the
difference of the non-equilibrium and equilibrium correlation function of the
fluctuations.
A spontaneous variation of the extensive variables x
i
will lead to a rate of
change of entropy given by
dS
K
dt
=

i
φ
i
dx
i
dt
. (11.63)
Using the property that for small deviations with respect to the steady state
φ
i
−φ
ss
i
=

j
_

2
S
∂x
i
∂x
j
_
ss
(x
j
−x
ss
j
), (11.64)
(11.54) can be written as
δ
2
S
K
=

i

i
−φ
ss
i
)(x
i
−x
ss
i
) (11.65)
and, after differentiation with respect to time,
d
dt
δ
2
S
K
= 2

i

i
−φ
ss
i
)
dx
i
dt
≥ 0. (11.66)
292 11 Mesoscopic Thermodynamic Descriptions
Rearranging (11.66), one obtains

i
φ
i
dx
i
dt

i
φ
ss
i
dx
i
dt
. (11.67)
The left-hand side of inequality (11.67) refers to the instantaneous values
of the intensive variables whereas the right-hand side involves their average
values in the steady state, but in virtue of (11.63), the left-hand side is also
the rate of change of entropy so that finally,
dS
K
dt

i
φ
ss
i
dx
i
dt
, (11.68)
which was called by Keizer a generalized Clausius inequality because it gen-
eralizes Clausius inequality T
R
dS/dt > dQ/dt established for a system in
equilibrium with a reservoir at temperature T
R
.
Keizer’s theory has been the subject of numerous applications as ion
transport through biological membranes, isomerization reactions, fluctuations
caused by electro-chemical reactions, light scattering under thermal gradients,
laser heated dimerization, etc. (Keizer 1987).
11.5 Mesoscopic Non-Equilibrium Thermodynamics
At short time and small length scales, the molecular nature of the systems
cannot be ignored. Classical irreversible thermodynamics(CIT) is no longer
satisfactory; indeed, molecular degrees of freedom that have not yet relaxed
to their equilibrium value will influence the global dynamics of the system,
and must be incorporated into the description, as done for example in ex-
tended thermodynamics or in internal variables theories. Unlike these ap-
proaches, which use as variables the average values of these quantities and
in contrast also with Keizer’s theory that adds, as additional variables, the
non-equilibrium part of the second moments of their fluctuations, mesoscopic
non-equilibrium thermodynamics describes the system through a probability
distribution function P(x, t), where x represents the set of all relevant de-
grees of freedom remaining active at the time and space scales of interest
(Reguera 2004; Rub´ı 2004).
The main idea underlying mesoscopic non-equilibrium thermodynamics is
to use the methods of CIT to obtain the evolution equation for P(x, t). The
selected variables do not refer to the microscopic properties of the molecules,
as for instance in the kinetic theory, but are obtained from an averaging pro-
cedure, as in macroscopic formulations. It is in this sense that the theory
is mesoscopic, i.e. intermediate between macroscopic and microscopic de-
scriptions. It shares some characteristics with extended thermodynamics or
internal variables theories, like an enlarged choice of variables and statistical
theories like the statistical concept of distribution function.
11.5 Mesoscopic Non-Equilibrium Thermodynamics 293
11.5.1 Brownian Motion with Inertia
As illustration, we will consider the role of inertial effects in the problem
of diffusion of N non-interacting Brownian particles of mass m immersed
in a fluid of volume V . Inertial effects are relevant when changes in spatial
density occur at timescales comparable to the time required by the velocity
distribution of particles to relax to its equilibrium Maxwellian value. At short
timescales, the particles do not have time to reach the equilibrium velocity
distribution, therefore the local equilibrium hypothesis cannot remain valid
and fluctuations become relevant. They are modelled by introducing as extra
variables the probability density P(r, v, t) to find the system with position
between r and r + dr and velocity between v and v + dv at time t. The
problem to be solved is to obtain the evolution equation of P(r, v, t).
The connection between entropy and probability of a state is given by the
Gibbs’ entropy postulate (e.g. de Groot and Mazur 1962), namely
s = s
eq

k
B
m
_
P(r, v, t) ln
P(r, v, t)
P
eq
(r, v)
dr dv, (11.69)
where s is the entropy per unit mass and P
eq
(r, v) the equilibrium distribu-
tion. By analogy with CIT, we formulate a Gibbs’ equation of the form
T ds = −
_
µ(r, v, t)dP(r, v, t)dr dv, (11.70)
where T is the temperature of a heat reservoir and µ the chemical potential
per unit mass, which can be given the general form
µ(r, v, t) = µ
eq
+
k
B
T
m
ln
P(r, v, t)
P
eq
(r, v)
+K(r, v), (11.71)
wherein K(r, v) is an extra potential which does not depend on P(r, v, t).
Since no confusion is possible between chemical potentials measured per unit
mass or per mole, we have omitted the horizontal bar surmounting “µ”. For
an ideal system of non-interacting particles in absence of external fields, sta-
tistical mechanics considerations suggest to identify K(r, v) with the kinetic
energy per unit mass
K(r, v) =
1
2
v
2
. (11.72)
The evolution equation for the distribution function will be given by the
continuity equation in the position and velocity space, namely
∂P
∂t
= −
∂J
r
∂r

∂J
v
∂v
, (11.73)
where J
r
and J
v
are the probability fluxes in the r, v space. The diffusion
flux of particles in the physical r coordinate space is directly obtained by
integration over the velocity space
294 11 Mesoscopic Thermodynamic Descriptions
¯
J
r
(r, t) =
_
vP(r, v, t)dv. (11.74)
The fluxes J
r
and J
v
in (11.73) are not known a priori, and will be derived
by following the methodology of non-equilibrium thermodynamic methods,
i.e. by imposing the restrictions placed by the second law. By differentiating
(11.70) with respect to time and using (11.73), it is found that the rate of
entropy production can be written as
T
ds
dt
=
_ _
−J
r

∂µ
∂r
−J
v

∂µ
∂v
_
dr dv, (11.75)
after performing partial integrations and supposing that the fluxes vanish
at the boundaries. As in classical irreversible thermodynamics, one assumes
linear relations between the fluxes J and the forces, so that
J
r
= −L
rr
∂µ
∂r
−L
rv
∂µ
∂v
, (11.76)
J
v
= −L
vr
∂µ
∂r
−L
vv
∂µ
∂v
, (11.77)
where L
ij
are phenomenological coefficients, to be interpreted later on. To
ensure the positiveness of the entropy production (11.75), the matrix of these
coefficients must be positive definite. Furthermore, if we take for granted the
Onsager–Casimir’s reciprocity relations, one has L
rv
= −L
vr
, with the minus
sign because r and v have opposite time-reversal parity.
To identify the phenomenological coefficients, substitute (11.71) in (11.76)
and impose the condition that the particle diffusion flux in the r-space should
be recovered from the flux in the r, v space, namely
¯
J
r
(r, t) =
_
vP(r, v, t)dv =
_
J
r
(r, v, t)dv
= −
_ _
L
rr
k
B
T
m
1
P
∂P
∂r
+L
rv
k
B
T
m
1
P
∂P
∂v
+L
rv
v
_
dv. (11.78)
Since P(r, v, t) is arbitrary, (11.78) may only be identically satisfied if L
rr
=0,
L
rv
= −P, and the only left undetermined coefficient is L
vv
. If it is taken as
L
vv
= P/τ, where τ is a velocity relaxation time related to the inertia of the
particles, (11.76) and (11.77) become, respectively,
J
r
=
_
v +
D
τ

∂v
_
P, (11.79)
J
v
= −
_
D
τ

∂r
+
v
τ
+
D
τ
2

∂v
_
P, (11.80)
where D ≡ (k
B
T/m)τ is identified as the diffusion coefficient. When these
expressions are introduced into the continuity equation (11.73), it is found
that
11.5 Mesoscopic Non-Equilibrium Thermodynamics 295
∂P
∂t
= −

∂r
(vP) +

∂v

_
v
τ
+
D
τ
2

∂v
_
P. (11.81)
This is the well-known Fokker–Planck’s equation for non-interacting Brownian
particles in presence of inertia and it is worth to stress that this equation has
been derived only by thermodynamic methods, without explicit reference to
statistical mechanics. Equilibrium situations corresponds to the vanishing of
the fluxes, i.e. J
r
= 0, J
v
= 0 and a Gaussian probability distribution. In
CIT, it is assumed that the probability distribution is the same as in equi-
librium, i.e. Gaussian, centred at a non-zero average, and with its variance
related to temperature in the same way as in equilibrium. Moreover, diffu-
sion in the coordinate r-space is much slower than in the velocity v-space,
so that it is justified to put J
v
= 0. From (11.80), it is then seen that
(v + (D/τ)∂/∂v)P = −D∂P/∂r which, substituted in (11.79), yields Fick’s
law
J
r
= −D
∂µ
∂r
. (11.82)
Far from equilibrium, neither the space nor the velocity distributions corre-
spond to equilibrium and one has J
r
,= 0, J
v
,= 0. In this case, the velocity
distribution may be very different from a Gaussian form, and it is not clear
how to define temperature (see, however, Box 11.1 where an attempt to define
a non-equilibrium temperature is presented).
Box 11.1 Non-Equilibrium Temperature
In the present formalism, it is not evident how to define a temperature
outside equilibrium. To circumvent the problem, a so-called effective tem-
perature has been introduced. It is defined as the temperature at which
the system is in equilibrium, i.e. the one corresponding to the probability
distribution at which the rate of entropy production vanishes. Substituting
(11.79) and (11.80) in (11.75), it is easily proven that the entropy produc-
tion can be written as
ds
dt
=
_
P

_
v +k
B
T

∂v
ln P
_
2
drdv, (11.1.1)
from that the effective temperature will be given by
1
T
eff
= −
k
B
v

∂v
ln P. (11.1.2)
This expression can be rewritten in a form recalling that of the equilibrium
temperature, i.e.
1
T
eff
=
∂s
eff
∂e
, (11.1.3)
at the condition to define an “effective” entropy by s
eff
= −k
B
ln P and an
energy density by e = (1/2)v
2
. Other definitions of effective temperature are
296 11 Mesoscopic Thermodynamic Descriptions
possible. Indeed, by taking e = (1/2)(v − ¸v))
2
, where ¸v
2
) is the average
value of v
2
, the temperature would be the local equilibrium one, but this
would not correspond to a zero entropy production. Moreover, as mentioned
by Vilar and Rub´ı (2001), the effective temperature is generally a function of
r, v, t; this means that, at a given position in space, there is no temperature
at which the system can be at equilibrium, because T(r, v, t) ,= T(r, t). If
it is wished to define a temperature at a position x, it would depend on the
way the additional degrees of freedom are eliminated.
One more remark is in form. It is rather natural to expect that the
mesoscopic theory discussed so far will cope with evolution equations of the
Maxwell–Cattaneo type, as the latter involve characteristic times compara-
ble to the relaxation time for the decay of the velocity distributions towards
its equilibrium value. Indeed by multiplying the Fokker–Planck’s equation
(11.81) by v and integrating over v, one obtains
τ

¯
J
∂t
= −
¯
J −D∇n, (11.83)
where
¯
J is the diffusion flux in space, given by (11.64), n(r, t) =
_
P(r, v, t)dv
the particle number density, and D = ¸v
2
)τ = (k
B
T/m) τ. This indicates
that mesoscopic non-equilibrium thermodynamics is well suited to describe
processes governed by relaxation equations of the Maxwell–Cattaneo type,
just like EIT. But now, it is the probability distribution function which is
elevated to the status of variable as basic variable rather than the average
value of the diffusion flux. It should be added that working in the frame of
EIT allows also obtaining the second moments of the fluctuations of J by
combining the expression of extended entropy with Einstein’s relation (11.5).
The above considerations could leave to picture that mesoscopic thermody-
namics is more general than EIT, however, when the second moments of
fluctuations are taken as variables, besides their average value, EIT provides
an interesting alternative more easier to deal with in practical situations.
11.5.2 Other Applications
The above results are directly generalized when more degrees of freedom
than r and v are present. The probability density will then be a function
of the whole set of degrees of freedom, denoted x, so that P = P(x, t). The
analysis performed so far can be repeated by replacing in all the mathematical
expressions the couple r, v by x. In particular, the continuity equation will
take the form
∂P
∂t
= −
∂J
∂x
, (11.84)
11.5 Mesoscopic Non-Equilibrium Thermodynamics 297
where J is given by the constitutive relation
J = −
L
T
∂µ
∂x
. (11.85)
After substitution of (11.85) in (11.84) and use of (11.71) for the chemical
potential, one obtains the kinetic equation
∂P
∂t
=

∂x
_
D
∂P
∂x
+
D
k
B
T
∂Φ
∂x
P
_
, (11.86)
which is a generalization of the Fokker–Planck’s equation (11.81), where now
Φ is not the kinetic energy as in (11.72) but it includes the potential energy
related to the internal degrees of freedom.
Extension to non-linear situations, as in chemical reactions, does not raise
much difficulty. If chemical processes are occurring at short timescales, they
will generally take place from an initial to a final state through intermediate
molecular configurations. Let the variable x characterize these intermediate
states. The chemical potential is still given by (11.71) in which K is, for
instance, a bistable potential whose wells correspond to the initial and fi-
nal states while the maximum represents the intermediate barrier. Such a
description is applicable to several problems as active processes, transport
through membranes, thermionic emission, adsorption, nucleation processes
(Reguera et al. 2005). Let us show, in particular, that mesoscopic thermody-
namics leads to a kinetic equation where the reaction rate satisfies the mass
action law. The linear constitutive law (11.85) is generalized in the form
J = −k
B
L
1
z
∂z
∂x
, (11.87)
where z ≡ exp(µ/k
B
T) is the fugacity; an equivalent expression is
J = −D
∂z
∂x
, (11.88)
where D ≡ k
B
L/z represents the diffusion coefficient. Assuming D constant
and integrating (11.88) from the initial state 1 to the final state 2 yields
¯
J ≡
_
2
1
J dx = −D(z
2
−z
1
) = −D
_
exp
µ
2
k
B
T
−exp
µ
1
k
B
T
_
, (11.89)
where
¯
J is the integrated rate. Expression (11.89) can alternatively be cast
in the more familiar form of a kinetic law
¯
J = K[1 −exp(−//k
B
T)], (11.90)
where K stands for Dexp(µ
1
/k
B
T) and / = µ
2
− µ
1
is the affinity of the
reaction. When µ
i
/k
B
T ¸ 1, one recovers from (11.90), the classical linear
phenomenological law of CIT (see Chap. 4)
298 11 Mesoscopic Thermodynamic Descriptions
¯
J =
D
k
B
T
/. (11.91)
Up to now, the theory has been applied to homogeneous systems char-
acterized by the absence of gradients of thermal and mechanical quantities.
The constraint of uniform temperature is now relaxed and the question is
how to incorporate thermal gradients in the formalism. For the problem of
Brownian motion, this is achieved by writing the Gibbs’ equation as follows:
ds =
1
T
du −
1
T
_
µdP dx dv, (11.92)
where u is the energy density. The corresponding constitutive equations are
now given by (Reguera et al. 2005)
J
q
= −L
TT
∇T/T
2

_
k
B
L
Tv

∂v
ln
P
P
leq
dv, (11.93)
J
v
= −L
vT
∇T/T
2

_
k
B
L
vv

∂v
ln
P
P
leq
dv, (11.94)
where J
q
is the heat flux, J
v
the probability current, and P
leq
the local equi-
librium distribution function; L
ij
are phenomenological coefficients forming
a positive definite matrix and obeying the Onsager relation L
Tv
= −L
vT
.
Equations (11.93) and (11.94) exhibit clearly the coupling between the two
irreversible processes occurring in the system: diffusion probability and heat
conduction. The corresponding evolution equation of the probability density
is now
∂P
∂t
= −v ∇P +β

∂v

_
Pv +k
B
T
∂P
∂v
_
+
γ
T

∂v
(P∇T), (11.95)
where β is the friction coefficient of the particles and γ a coefficient related
to L
vT
.
To summarize, mesoscopic non-equilibrium thermodynamics has shown
its applicability in a wide variety of situations where local equilibrium is
never reached, as for instance relaxation of polymers or glasses, dynamics of
colloids or flows of granular media. Such systems are characterized by internal
variables exhibiting short length scales and slow relaxation times. The basic
idea of this new theory is to incorporate in the description those variables,
which have not yet relaxed to their local equilibrium values. As illustrative
examples, we have discussed the problem of inertial effects in diffusion of
Brownian particles where change of density take place at a timescale of the
order of the time needed by the particles to relax to equilibrium, chemical
reactions and thermodiffusion. The tools are borrowed from CIT (see Chap. 2)
but the original contribution is to introduce, amongst the set of variables, the
density probability distribution whose time evolution is shown to be governed
by a Fokker–Planck’s equation.
11.6 Problems 299
11.6 Problems
11.1. Second moments. (a) Show, from (11.12) and (11.15) that the second
moments of the energy and the volume fluctuations are given by
¸δUδU) = −(k
B
/M)(∂U/∂T
−1
)
T
−1
p
,
¸δUδV ) = −(k
B
/M)(∂V/∂T
−1
)
T
−1
p
,
¸δV δV ) = −(k
B
/M)(∂V/∂T
−1
p)
T
−1.
(b) Write explicitly the second-order derivatives appearing in (11.12) and
(11.15), and derive expressions (11.16).
11.2. Density fluctuations. (a) From the second moments of the volume fluc-
tuations, write the expression for the density fluctuations of a one-component
ideal gas at 0

C and 1 atm, when the root-mean-square deviation in density
is 1% of the average density of the system? (b) Show that near a critical
point, where (∂p/∂V )
T
= 0, these fluctuations diverge.
11.3. Dielectric constant. The dielectric constant ε of a fluid varies with the
mass density according to the Clausius–Mossoti’s relation
ε −1
ε + 2
= Cρ,
with C is a constant related to the polarizability of the molecules and ρ is
the mass density. Show that the second moments of the fluctuations of ε are
given by
¸(δε)
2
) =
k
B

T
9V
(ε −1)
2
(ε + 2)
2
,
with κ
T
being the isothermal compressibility.
11.4. Density fluctuations and non-locality. The correlations in density fluc-
tuations at different positions are usually written as ¸δn(r
1
)δn(r
2
)) ≡
¯ nδ(r
1
− r
2
) + ¯ nν(r) with ¯ n the average value of the density, r ≡ [r
2
−r
1
[,
and ν(r) the correlation function. To describe such correlations, Ginzburg
and Landau propose to include in the free energy a non-local term of the
form
F(T, n, ∇n) = F
eq
(T, n) −
1
2
b(∇n)
2
=
1
2
a(n − ¯ n)
2

1
2
b(∇n)
2
,
with a(T) a function of temperature which vanishes at the critical point and
b > 0 a positive constant. (a) If the density of fluctuations are expressed as
n − ¯ n =

k
n
k
e
ik·r
,
300 11 Mesoscopic Thermodynamic Descriptions
show that
¸[δn
k
[
2
) =
k
B
T
V (a +bk
2
)
.
(b) Taking into account that
_
ν(r)e
−ik·r
dV =
V
n
¸[δn
k
[
2
) −1,
prove that
ν(r) =
k
B
Ta
4π¯ nb
1
r
exp
_

r
ξ
_
,
where ξ is a correlation length, given by ξ = (b/a)
1/2
. (Note that in the limit
ξ →0, one obtains ν(r) = δ(r), with δ(r) the Dirac’s function, and that near
a critical point the correlation length diverges.)
11.5. Transport coefficients. (a) Apply (11.43) to obtain the classical results
η = nk
B
Tτ and λ = (5k
B
T
2
/2m)τ. In terms of the peculiar molecular ve-
locities c the microscopic operators for the fluxes are given by
ˆ
P
v
12
= mc
1
c
2
and ˆ q
1
=
_
1
2
mc
2

5
2
k
B
T
_
c
1
. The equilibrium average should be performed
over the Maxwell–Boltzmann distribution function. (b) Make a similar analy-
sis for the memory kernel corresponding to the diffusion coefficient, namely
D(t −t

) = (V/k
B
T) ¸δJ
i
(t)δJ
i
(t

)).
11.6. Second moments and EIT. Apply Einstein’s relation (11.5) to the en-
tropy (7.61) of EIT, and find the second moments of the fluctuations of the
fluxes around equilibrium. Note that the results coincide with (11.43), ob-
tained from Green–Kubo’s relations by assuming an exponential relaxation
of the fluctuations of the fluxes.
11.7. Fluctuations around steady states. Assume with Keizer that Einstein’s
relation remains valid around a non-equilibrium steady state. Determine the
second moments of the fluctuations of u and q around a non-equilibrium
steady state characterized by a non-vanishing average heat flux q
0
. Compare
the results with those obtained in equilibrium in Problem 11.6.
11.8. Brownian motion. Langevin proposed to model the Brownian motion
of the particles by adding to the hydrodynamic friction force −ζv (ζ is the
friction coefficient, v is the speed of the particle), a stochastic force f de-
scribing the erratic forces due to the collisions of the microscopic particles of
the solvent, in such a way that
m
dv
dt
= −ζv +f .
He assumed that f is white (without memory) and Gaussian, in such a way
that ¸f ) = 0, ¸f (t)f (t +t

)) = Bδ(t

). Using the result that in the long-time
limit the equipartition condition ¸
1
2
mv
2
) =
3
2
k
B
T must be satisfied, show
11.6 Problems 301
that ¸f (t)f (t +t

)) = (ζ/m)k
B
Tδ(t

), i.e. B = (ζ/m)k
B
T. This is similar to a
fluctuation–dissipation relation, as it shows that the second moments of the
fluctuating force are proportional to the friction coefficient. This is called a
fluctuation–dissipation relation of the second kind.
11.9. Generalized fluctuation–dissipation relation. The result of the previous
problem may be written in a more general way for any set a(t) of random
variables satisfying a linear equation of the form da/dt + H a = f with
H a friction matrix and f a white noise such that ¸f ) = 0, ¸f (t)f
T
(t +
t

)) = Bδ(t

). Show that the matrix Bmust satisfy the fluctuation–dissipation
relation
H G+G H
T
= B,
with G ≡ ¸δa δa
T
) the matrix of the second moments of fluctuations of a(t).
11.10. Non-equilibrium temperature. To underline the connection between
the non-equilibrium temperature in Keizer’s formalism and in EIT, note that,
when the heat flux q is the only relevant flux, EIT predicts that ¸δuδu) =
¸δuδu)
eq

q
2
, where subscript eq mean local equilibrium. In this expression
δu is the fluctuations of the internal energy with respect to its steady state
average and α

a coefficient, whose explicit form is given in Jou et al. (2001).
The above result can still be cast in the form
q
2
= [¸δuδu) −¸δuδu)
eq
] (α

)
−1
.
Recalling that in EIT the non-equilibrium temperature T is given by
1
T
=
1
T
eq

1
2
∂α
∂u
q
2
,
write this expression in terms of the second moments of energy fluctuations
and compare with Keizer’s expression (11.60).
Epilogue
By writing this book, our objective was threefold:
1. First, to go beyond equilibrium thermodynamics. Although it is widely
recognized that equilibrium thermodynamics is a universal and well-
founded discipline with many applications mainly in chemistry and en-
gineering, it should be realized that its domain of application is limited
to equilibrium states and idealized reversible processes, excluding dissi-
pation. This is sufficient to predict the final equilibrium state, knowing
the initial state, but it is silent about the duration and the nature of the
actual process between the initial and the final equilibrium states, whence
the need to go beyond equilibrium thermodynamics. Another reason is
that, to foster the contact between micro- and macroscopic approaches, it
is imperative to go beyond equilibrium as most of the microscopic theories
deal with situations far from equilibrium. But the problem we are faced
with is that the avenue of equilibrium thermodynamics bifurcates in many
routes.
2. Our second objective was to propose a survey, as complete as possible,
of the many faces of non-equilibrium thermodynamics. For pedagogical
reasons, we have restricted the analysis to the simplest situations, empha-
sizing physical rather than mathematical aspects. The presentation of each
theory is closed by a critical discussion from which can be concluded that
none of the various approaches is fully satisfactory. It appears that each
school has its own virtues but that, in practical situations, one of them
may be preferable to another. It is our purpose that after gone through the
present book, the reader will be able to make up his personal opinion. We
do not pretend to have been everywhere fully objective and exhaustive.
We have deliberately been silent about some valuable descriptions, as the
entropy-free theory of Meixner (1973a, b), the Lagrangian formalism of
Biot (1970), the variational analysis of Sieniutycz (1994), the thermody-
namics of chaos (Beck and Schl¨ogl 1993; Berdichevsky 1997; Gaspard 1998;
Ruelle 1991), the statistical approach of Luzzi et al. (2001, 2002), and
303
304 Epilogue
Tsallis’ non-extensive entropy formalism (2004). For reasons of place and
unity, and despite their intrinsic interest, we have also deliberately omitted
microscopic formulations, such as statistical mechanics (e.g. Grandy 1987),
kinetic theory (e.g. Chapman and Cowling 1970), information theory
(Jaynes 1963), molecular dynamics (Evans and Morriss 1990), and other
kinds of computer simulations (e.g. Hoover 1999; Hutter and J¨ ohnk 2004).
We apologize for these omissions; the main reason was our option to confine
the volume of the book to a reasonable size rather than to write an exten-
sive encyclopaedia. We have also bypassed some approaches either because
of their more limited impact in the scientific community, or because they
lack of sufficiently new fundamental ideas or techniques. Concerning this
multiplicity, it may be asked why so many thermodynamics? A tentative
answer may be found in the diversity of thought of individuals, depending
on their roots, environment, and prior formation as physicists, mathe-
maticians, chemists, engineers, or biologists. The various thermodynamic
theories are based on different foundations: macroscopic equilibrium ther-
modynamics, kinetic theory, statistical mechanics, or information theory.
Other causes of diversity may be found in the selection of the most
relevant variables and the difficulty to propose an undisputed defini-
tion of temperature, entropy, and the second law outside equilibrium.
At the exception of classical irreversible thermodynamics (CIT), it is
generally admitted that non-equilibrium entropy depends, besides clas-
sical quantities as mass, energy, charge density, etc., on extra variables
taking the form of dissipative fluxes in extended irreversible thermody-
namics (EIT), internal structural variables in internal variables theories
(IVT) and in GENERIC and probality distribution function in meso-
scopic theories. Out of equilibrium, the constitutive relations can either
be cast in the form of linear algebraic phenomenological relations as in
CIT, integrals involving the memory as in rational thermodynamics (RT),
or time evolution differential equations as in EIT, IVT, and GENERIC.
The next natural question is then: what is the best approach? Although the
present authors have their own (subjective) opinion, we believe that the
final answer should be left to each individual reader but there is no doubt
that trying to reach unanimity remains a tremendous challenging task.
3. We failed to meet a third objective, namely to bring a complete unity
into non-equilibrium thermodynamics. Being aware about the role of non-
uniformity and the importance of diversity, we realize that the achievement
of such a unity may appear as illusory. However, we do not think that it
is a completely desperate task; indeed, it is more than a dream to believe
that in a near future it would be possible to summon up all the pieces
of the puzzle and to build up a well-shaped, unique, and universal non-
equilibrium thermodynamics. In that respect, we would like to stress that
there exists a wide overlapping between the different schools. More specif-
ically, all the theories contain as a special case the classical irreversible
thermodynamics. The Cattaneo model of heat conduction is not typical of
Epilogue 305
EIT but can also be obtained in the framework of IVT, RT, GENERIC,
and the mesoscopic description. It is our hope that the present book will
contribute to promote reconciliation among the several approaches and
foster further developments towards deeper and more unified formulations
of thermodynamics beyond equilibrium.
By the way, it was also our purpose to convince the reader that “thermo-
dynamics is the science of everything”. Clearly, thermodynamics represents
more than converting heat into work or calculating engine efficiencies. It is
a multi-disciplinary science covering a wide variety of fields ranging from
thermal engineering, fluid and solid mechanics, rheology, material science,
chemistry, biology, electromagnetism, cosmology to economical, and even so-
cial sciences.
Among the several open and challenging problems, let us mention three of
them. The first one is related to the limits of applicability of thermodynamics
to small systems, like found in nano-technology and molecular biophysics. At
the opposite, in presence of long-range interactions, such as gravitation, it
may be asked how the second law should be formulated when these effects
are dominant, like in cosmology. Finally, does thermodynamics conflict with
quantum mechanics? how to reconcile the reversible laws of quantum the-
ory and the subtleties of quantum entanglement of distant systems with the
irreversible nature of thermodynamics?
By writing this book, we were guided by the intellectual ambition to better
understand the frontiers and perspectives of the multi-faced and continuously
changing domain of our knowledge in non-equilibrium thermodynamics. This
remains clearly an unfinished task and we would like to think to have con-
vinced the reader that, therefore, this fascinating story is far from
The End
References
Agrawal D.C. and Menon V.J., Performance of a Carnot refrigerator at maximum
cooling power, J. Phys. A: Math. Gen. 23 (1990) 5319–5326
Alvarez X. and Jou D., Memory and nonlocal effects in heat transport: from diffusive
to ballistic regimes, Appl. Phys. Lett. 90 (2007) 083109 (3 pp)
Andresen B., Finite-time thermodynamics in simulated annealing, in Entropy and
Entropy Generation (Shiner J.S., ed.), Kluwer, Dordrecht, 1996
Angulo-Brown F., An ecological optimization criterion for finite-time heat engines,
J. Appl. Phys. 69 (1991) 7465–7469
Anile A.M. and Muscato O., An improved thermodynamic model for carrier transport
in semiconductors, Phys. Rev. B 51 (1995) 16728–16740
Anile A.M. and Pennisi S., Extended thermodynamics of the Blotekjaer hydrodynam-
ical model for semiconductors, Continuum Mech. Thermodyn. 4 (1992) 187–197
Anile A.M., Allegretto W. and Ringhofer C., Mathematical Problems in Semi-
Conductor Physics, Lecture Notes in Mathematics, vol. 1823, Springer, Berlin
Heidelberg New York, 2003
Astumian R.D., Thermodynamics and kinetics of a Brownian motor, Science 276
(1997) 917–922
Astumian R.D. and Bier M., Fluctuation driven ratchets. Molecular motors, Phys.
Rev. Lett. 72 (1994) 1766–1769
Bampi F. and Morro A., Nonequilibrium thermodynamics: a hidden variable
approach, in Recent Developments in Non-Equilibrium Thermodynamics (Casas-
V´ azquez J., Jou D. and Lebon G., eds.), Lecture Notes in Physics, vol. 199,
Springer, Berlin Heidelberg New York, 1984
Beck C. and Schl¨ogl F., Thermodynamics of Chaotic Systems, Cambridge University
Press, Cambridge, 1993
Bedeaux D., Mazur P. and Pasmanter R.A., The ballast resistor; an electro-thermal
instability in a conducting wire I; the nature of the stationary states, Physica A
86 (1977) 355–382
Bejan A., Entropy Generation Minimization, CRC, Boca Raton, FL, 1996
B´enard H., Les tourbillons cellulaires dans une nappe liquide transportant la chaleur
par conduction en r´egime permanent, Rev. G´en. Sci. Pures et Appliqu´ees 11 (1900)
1261–1271, 1309–1318
Berdichevsky V.L., Thermodynamics of Chaos and Order, Longman, Essex, 1997
Berg´e P., Pomeau Y. and Vidal C., L’ordre dans le chaos, Hermann, Paris, 1984
Beris A.N., Simple Non-equilibrium thermodynamics application to polymer rheology,
in Rheology Reviews (Binding D.N. and Walters K., eds.), pp 33–75, British Society
of Rheology, Aberystwyth, 2003
307
308 References
Beris A.N. and Edwards B.J., Thermodynamics of Flowing Systems with Internal
Microstructure, Oxford Science Pubisher, Oxford, 1994
Biot M.A., Variational Principles in Heat Transfer, Oxford University Press, Oxford,
1970
Bird R.B. and de Gennes P.G., Discussion about the principle of objectivity, Physica
A 118 (1983) 43–47
Bird R.B., Armstrong R.C. and Hassager D., Dynamics of Polymeric Liquids, 2nd
ed., vol. 1: Fluid Mechanics, Wiley, New York, 1987a
Bird R.B., Curtiss C.F., Armstrong R.C. and Hassager D., Dynamics of Polymeric
Liquids, 2nd ed., vol. 2: Kinetic Theory, Wiley, New York, 1987b
Block M.J., Surface tension as the cause of B´enard’s cells and surface deformation in
a liquid film, Nature 178 (1956) 650
Bodenshatz E., Pech W. and Ahlers G., Recent developments in Rayleigh-B´enard
convection, Annu. Rev. Fluid Mech. 32 (2000) 851–919
Bragard J. and Lebon G., Nonlinear Marangoni convection in a layer of finite depth,
Europhys. Lett. 21 (1993) 831–836
Braun D. and Libchaber A., Phys. Rev. Lett. 89 (2002) 188103
Brenner H., Rheology of two-phase systems, Annu. Rev. Fluid Mech. 2 (1970) 137–176
Bricmont J., Science of chaos or chaos in science? Physicalia Mag. 17 (1995) 159–221
Bridgman P.W., The Nature of Thermodynamics, Harvard University Press,
Cambridge, 1941
Busse F.H., Non-linear properties of thermal convection, Rep. Prog. Phys. 41 (1978)
1929–1967
Bustamante C., Liphardt J. and Ritort F., The nonequilibrium thermodynamics of
small systems, Phys. Today 58 (2005) 43–48
Callen H.B., Thermodynamics and an Introduction to Thermostatistics, 2nd ed.,
Wiley, New York, 1985
Callen H.B. and Welton T.A., Irreversibility and generalized noise, Phys. Rev. 83
(1951) 34–40
Callen H.B. and Greene R.F., On a theorem of irreversible thermodynamics, Phys.
Rev. 86 (1952) 702–706
Camacho J. and Zakari M., Irreversible thermodynamic analysis of a two-layer system,
Phys. Rev. E 50 (1994) 4233–4236
Caplan S.R. and Essig A., Bioenergetics and Linear Nonequilibrium Thermodynam-
ics. The Steady State, Harvard University Press, Cambridge, 1983
Carrassi M. and Morro A., A modified Navier–Stokes equations and its consequences
on sound dispersion, Nuovo Cimento B 9 (1972) 321–343; 13 (1984) 281–289
Casas-V´azquez J. and Jou D., Temperature in non-equilibrium states: a review of
open problems and current proposals, Rep. Prog. Phys. 66 (2003) 1937–2023
Casas-V´azquez J., Jou D. and Lebon G. (eds.), Recent Developments in Non-
equilibrium Thermodynamics, Lecture Notes in Physics, vol. 199, Springer, Berlin
Heidelberg New York, 1984
Casimir H.B.G., On Onsager’s principle of microscopic irreversibility, Rev. Mod. Phys.
17 (1945) 343–350
Cattaneo C., Sulla conduzione del calore, Atti Seminario Mat. Fis. University Modena
3 (1948) 83–101
Chandrasekhar S., Hydrodynamic and Hydromagnetic Stability, Clarendon, Oxford,
1961
Chapman S. and Cowling T.G., The Mathematical Theory of Non-Uniform Gases,
Cambridge University Press, Cambridge, 1970
Chartier P., Gross M. and Spiegler K.S., Applications de la thermodynamique du
non-´equilibre, Hermann, Paris, 1975 (in French)
Chen G., Nanoscale Heat Transfer and Conversion, Oxford University Press, Oxford,
2004
References 309
Cimmelli V.A., Boundary conditions in the presence of internal variables, J. Non-
Equilib. Thermodyn. 27 (2002) 327–348
Cloot A. and Lebon G., A nonlinear stability analysis of the B´enard–Marangoni
problem, J. Fluid Mech. 45 (1984) 447–469
Coleman B.D., Thermodynamics of materials with memory, Arch. Rat. Mech. Anal.
17 (1964) 1–46
Coleman B.D. and Owen D., A mathematical foundation of thermodynamics, Arch.
Rat. Mech. Anal. 54 (1974) 1–104
Coleman B.D. and Owen D., On thermodynamics and elastic–plastic materials, Arch.
Rat. Mech. Anal. 59 (1975) 25–51
Coleman B.D. and Truesdell C., On the reciprocal relations of Onsager, J. Chem.
Phys. 33 (1960) 28–31
Coleman B.D., Owen D. and Serrin J., The second law of thermodynamics for systems
with approximate cycles, Arch. Rat. Mech. Anal. 77 (1981) 103–142
Coles D., Transition in circular Couette flow, J. Fluid Mech. 21 (1965) 385–425
Colinet P., Legros J.C. and Velarde M., Nonlinear Dynamics of Surface-Tension
Driven Instabilities, Wiley, Berlin, 2002
Criado-Sancho M. and Casas-V´azquez J., Termodin´ amica qu´ımica y de los procesos
irreversibles, 2nd ed., Pearson/Addison-Wesley, Madrid, 2004 (in Spanish)
Criado-Sancho M. and Llebot J.E., Behaviour of entropy in hyperbolic heat transport,
Phys. Rev. E 47 (1993) 4104–4108
Crisanti A. and Ritort F., Violation of the fluctuation–dissipation theorem in glassy
systems: basic notions and the numerical evidence, J. Phys. A: Math. Gen. 36
(2003) R181–R290
Cross M.C. and Hohenberg P.C., Pattern formation outside equilibrium, Rev. Mod.
Phys. 65 (1993) 852–1112
Curzon F.L. and Ahlborn B., Efficiency of a Carnot engine at maximum power output,
Am. J. Phys. 43 (1975) 22–24
Dauby P.C. and Lebon G., B´enard–Marangoni instability in rigid rectangular con-
tainers, J. Fluid Mech. 329 (1996) 25–64
Dauby P.C., Bouhy E. and Lebon G., Linear B´enard–Marangoni instability in rigid
circular containers, Phys. Rev. E 56 (1997) 520–530
Davis S.H., Buoyancy-surface tension instability by the method of energy, J. Fluid
Mech. 39 (1969) 347–359
Davis S.H., Thermocapillary instabilities, Annu. Rev. Fluid Mech. 19 (1987) 403–435
Day W., The Thermodynamics of Simple Materials with Fading Memory, Springer,
Berlin Heidelberg New York, 1972
Day W., An objection against using entropy as a primitive concept in continuum
thermodynamics, Acta Mechanica 27 (1977) 251–255
Dedeurwaerdere T., Casas-V´azquez J., Jou D. and Lebon G., Foundations and appli-
cations of a mesoscopic thermodynamic theory of fast phenomena, Phys. Rev. E
53 (1996) 498–506
Doi M. and Edwards S.F., The Theory of Polymer Dynamics, Clarendon, Oxford,
1986
Drazin P.G. and Reid W.H., Hydrodynamic Stability, Cambridge University Press,
Cambridge, 1981
Dreyer W. and Struchtrup H., Heat pulse experiments revisited, Continuum Mech.
Thermodyn. 5 (1993) 3–50
Duhem P., Trait´e d’´energ´etique ou de thermodynamique g´en´erale, Gauthier-Villars,
Paris, 1911
Eckart C., The thermodynamics of irreversible processes. I–III, Phys. Rev. 58 (1940)
267–269, 269–275, 919–924; IV, Phys. Rev. 73 (1948) 373–382
Edelen D.G.D. and McLennan J.A., Material indifference: a principle or a conve-
nience, Int. J. Eng. Sci. 11 (1973) 813–817
310 References
Elmer F.J., Limit cycle of the ballast resistor caused by intrinsic instabilities, Z. Phys.
B 87 (1992) 377–381
Eringen C., Mechanics of Continua, Wiley, New York, 1967
Eu B.C., Kinetic Theory and Irreversible Thermodynamics, Wiley, New York, 1992
Eu B.C., Nonequilibrium Statistical Mechanics: Ensemble Method, Kluwer,
Dordrecht, 1998
Evans D.J. and Morriss G.P., Statistical Mechanics of Non-Equilibrium Liquids,
Academic, London, 1990
Finlayson B., The Method of Weighted Residuals and Variational Principles,
Academic, New York, 1972
Fowler R.H. and Guggenheim E.A., Statistical Thermodynamics, Cambridge
University Press, Cambridge, 1939
Frankel N.A. and Acrivos A., The constitutive equations for a dilute emulsion,
J. Fluid Mech. 44 (1970) 65–78
Garc´ıa-Colin L.S., Extended irreversible thermodynamics beyond the linear regime:
a critical overview, J. Non-Equilib. Thermodyn. 16 (1991) 89–128
Garcia-Colin L.S., Extended irreversible thermodynamics: an unfinished task, Mol.
Phys. 86 (1995) 697–706
Gaspard P., Chaos, Scattering and Statistical Mechanics, Cambridge University
Press, Cambridge, 1998
Gibbs J.W., The Collected Works of J.W. Gibbs, vol. 1: Thermodynamics, Yale
University Press, Yale, 1948
Glansdorff P. and Prigogine I., On a general evolution criterion in macroscopic
physics, Physica 30 (1964) 351–374
Glansdorff P. and Prigogine I., Thermodynamics of Structure, Stability and Fluctu-
ations, Wiley, New York, 1971
Gollub J.P. and Langer J.S., Pattern formation in nonequilibrium physics, Rev. Mod.
Phys. 71 (1999) S396–S403
Gorban A.N., Karlin I.V. and Zinovyev A.Y., Constructive methods of invariant
manifolds for kinetic problems, Phys. Rep. 396 (2004) 197–403
Gordon J.M. and Zarmi Y., Wind energy a solar-driven heat engine: a thermodynamic
approach, Am. J. Phys. 57 (1989) 995–998
Grad H., Principles of the Kinetic Theory of Gases, in Hd. der Physik, vol. XII (Flugge
S., ed.), Springer, Berlin Heidelberg New York, 1958
Grandy W.T., Foundations of Statistical Mechanics, Reidel, Dordrecht, 1987
Grmela M., Complex fluids subjected to external influences, J. Non-Newtonian Fluid
Mech. 96 (2001) 221–254
Grmela M. and Lebon G., Hamiltonian extended thermodynamics, J. Phys. A: Math.
Gen. 23 (1990) 3341–3351
Grmela M. and
¨
Ottinger H.C., Dynamics and thermodynamics of complex fluids. I.
Development of a generic formalism, Phys. Rev. E 56 (1997) 6620–6632
Grmela M., Elafif A. and Lebon G., Isothermal non-standard diffusion in a two-
component fluid mixture: a Hamiltonian approach, J. Non-Equilib. Thermodyn.
23 (1998) 312–327
Grmela M., Lebon G. and Lhuillier D., A comparative study of the coupling of flow
with non-Fickean thermodiffusion. II. Generic, J. Non-Equilib. Thermodyn. 28
(2003) 23–50
Grmela M., Lebon G., Dauby P.C. and Bousmina M., Ballistic-diffusive heat conduc-
tion at nanoscale: a generic approach, Phys. Lett. A 339 (2005) 237–245
de Groot S.R. and Mazur P., Non-Equilibrium Thermodynamics, North-Holland,
Amsterdam, 1962 (republication by Dover in 1984)
Guyer R.A. and Krumhansl J.A., Solution of the linearized Boltzmann phonon equa-
tion, Phys. Rev. 148 (1966) 766–788
References 311
Gyarmati I., Non-equilibrium Thermodynamics, Springer, Berlin Heidelberg New
York, 1970
Hafskold B. and Kjelstrup S., Criteria for local equilibrium in a system with transport
of heat and mass, J. Stat. Phys. 78 (1995) 463–494
Haken H. (ed.), Synergetics: A Workshop, Springer, Berlin Heidelberg New York,
1977
H¨ ansch W., The Drift-Diffusion Equation and Its Application in MOSFET Modeling,
Springer, Berlin Heidelberg New York, 1991
Hill T.L., Free Energy Transduction in Biology, the Steady-State Kinetic and Ther-
modynamic Formalism, Academic, New York, 1977
Hoffmann K.H., Burzler J.M. and Schubert S., Endoreversible thermodynamics,
J. Non-Equilib. Thermodyn. 22 (1997) 311–355
Hoover W.H., Time Reversibility, Computer Simulations and Chaos, World Scientific,
Singapore, 1999
Hoover W., Moran B., More R. and Ladd A., Heat conduction in a rotating disk via
non-equilibrium molecular dynamics, Phys. Rev. A 24 (1981) 2109–2115
Hutter K. and J¨ohnk K., Continuum Methods of Physical Modelling, Springer, Berlin
Heidelberg New York, 2004
Ichiyanagi M., Variational principles in irreversible processes, Phys. Rep. 243 (1994)
125–182
Ilg P. and
¨
Ottinger H.C., Non-equilibrium relativistic thermodynamics in bulk viscous
cosmology, Phys. Rev. D 61 (1999) 023510, 1–10
Jaynes E.T., in Statistical Physics (Ford W.K., ed.) Benjamin, New York, 1963
Joseph D.D., Stability of Fluid Motions, Springer, Berlin Heidelberg New York, 1976
Joseph D.D. and Preziosi L., Heat waves, Rev. Mod. Phys. 61 (1989) 41–74; 62 (1990)
375–392
Jou D. and Llebot J.E., Introduction to the Thermodynamics of Biological Processes,
Prentice Hall, New York, 1990
Jou D., Casas-V´azquez J. and Criado-Sancho M., Thermodynamics of Fluids Under
Flow, Springer, Berlin Heidelberg New York, 2000
Jou D., Casas-V´azquez J. and Lebon G., Extended Irreversible Thermodynamics, 3rd
ed., Springer, Berlin Heidelberg New York, 2001
Jou D., Casas-V´azquez J., Lebon G. and Grmela M., A phenomenological scaling
approach for heat transport in nano-systems, Appl. Math. Lett. 18 (2005) 963–
967
J¨ ulicher A., Ajdari A. and Prost P., Modeling molecular motors, Rev. Mod. Phys. 69
(1997) 1269–1282
Katchalsky A. and Curran P.F., Non-Equilibrium Thermodynamics in Biophysics,
Harvard University Press, Cambridge, 1965
Keizer J., Thermodynamics of non-equilibrium steady states, J. Chem. Phys. 69
(1978) 2609–2620
Keizer J, On the relationship between fluctuating irreversible thermodynamics and
‘extended’ irreversible thermodynamics, J. Stat. Phys. 31 (1983) 485–497
Keizer J., Statistical Thermodynamics of Nonequilibrium Processes, Springer, Berlin
Heidelberg New York, 1987
Kestin J., A Course in Thermodynamics, Blaisdell, Waltham, 1968
Kestin J., A note on the relation between the hypothesis of local equilibrium and the
Clausius–Duhem inequality, J. Non-Equilib. Thermodyn. 15 (1990) 193–212
Kestin J., Local-equilibrium formalism applied to mechanics of solids, Int. J. Solids
Struct. 29 (1992) 1827–1836
Kestin J. and Bataille J., Thermodynamics of Solids, in Continuum Models of Dis-
crete Systems, University of Waterloo Press, Waterloo, 1980
Khalatnikov I.M., An introduction to the Theory of Superfluidity, W.A. Benjamin,
New York, 1965
312 References
Kleidon A. and Lorenz R.D. (eds.), Non-Equilibrium Thermodynamics and the
Production of Entropy. Life, Earths and Beyond, Springer, Berlin Heidelberg New
York, 2005
Kluitenberg G., Thermodynamical theory of elasticity and plasticity, Physica 28
(1962) 217–232
Kondepudi D. and Prigogine I., Modern Thermodynamics: from Heat Engines to
Dissipative Structures, Wiley, New York, 1998
Koschmieder E.L., B´enard cells and Taylor Vortices, Cambridge University Press,
Cambridge, 1993
Kratochvil J. and Silhavy M., On thermodynamics of non-equilibrium processes,
J. Non-Equilib. Thermodyn. 7 (1982) 339–354
Kreuzer H.J., Non-Equilibrium Thermodynamics and Its Statistical Foundations,
Clarendon, Oxford, 1981
Kubo R., Statistical Mechanics, Wiley, New York, 1965
Landau L.D., Collected Papers (Haar T., ed.), Pergamon, Oxford, 1965
Landau L.D. and Lifshitz E.M., Statistical Physics, 3rd ed., Pergamon, Oxford, 1980
Landsberg P.T. and Tonge G., Thermodynamic energy conversion efficiencies, J. Appl.
Phys. 51 (1980) R1–R18
Lavenda B., Thermodynamics of Irreversible Processes, McMillan, London, 1979
Lebon G., Variational Principles in Thermomechanics, in CISME Courses and Lec-
ture Notes, vol. 262, Springer, Berlin Heidelberg New York, 1980
Lebon G. and Jou D., Linear irreversible thermodynamics and the phenomenological
theory of liquid helium II, J. Phys. C 16 (1983) L199–L204
Lebon G. and Dauby P.C., Heat transport in dielectric crystals at low temperature:
a variational formulation based on extended irreversible thermodynamics, Phys.
Rev. A 42 (1990) 4710–4715
Lebon G., Extended Thermodynamics, in CISME Courses and Lecture Notes, vol.
336, pp 139–204, Springer, Berlin Heidelberg New York, 1992
Lebon G., Casas-Vazquez J. and Jou D., Questions and answers about a thermody-
namic theory of the third type, Contemp. Phys. 33 (1992) 41–51
Lebon G., Torrissi M. and Valenti A., A nonlocal thermodynamic analysis of second
sound propagation in crystalline dielectrics, J. Phys. C 7 (1995) 1461–1474
Lebon G., Grmela M. and Lhuillier D., A comparative study of the coupling of
flow with non-Fickian thermodiffusion. I. Extended irreversible thermodynamics,
J. Non-Equilib. Thermodyn. 28 (2003) 1–22
Lebon G., Desaive Th. and Dauby P.C., A unified extended thermodynamic descrip-
tion of diffusion, thermo-diffusion, suspensions, and porous media, J. Appl. Mech.
73 (2006a) 16–20
Lebon G., Desaive Th. and Dauby P.C., Two-fluid diffusion as a model for heat
transport at micro and nanoscales, in Proceedings of 7th International Meeting on
Thermodiffusion, San Sebastian, 2006b
Lebon G., Lhuillier D. and Palumbo A., A thermodynamic description of thermodif-
fusion in suspensions of rigid particles, Eur. Phys. J.IV 146 (2007) 3–12
Lebowitz J.L., Microscopic origins of irreversible macroscopic behaviour, Physica A
263 (1999) 516–527
Lepri S., Livi R. and Politi A., Thermal conduction in classical low-dimensional lat-
tice, Phys. Rep. 377 (2003) 1–80
Lhuillier D., From molecular mixtures to suspensions of particles, J. Phys. II France
5 (1995) 19–36
Lhuillier D., Internal variables and the non-equilibrium thermodynamics of colloidal
suspensions, J. Non-Newtonian Fluid Mech. 96 (2001) 19–30
Lhuillier D. and Ouibrahim A., A thermodynamic model for solutions of deformable
molecules, J. M´ecanique 19 (1980) 725–741
References 313
Lhuillier D., Grmela M. and Lebon G., A comparative study of the coupling of
flow with non-Fickian thermodiffusion. Part III. internal variables, J. Non-Equilib.
Thermodyn. 28 (2003) 51–68
Liu I.-S., Method of Lagrange multipliers for exploitation of the entropy principle,
Arch. Rat. Mech. Anal. 46 (1972) 131–148
Lorenz E.N., Deterministic nonperiodic flow, J. Atmos. Sci. 20 (1963) 130–141
Lumley J., Turbulence modelling, J. Appl. Mech. 50 (1983) 1097–1103
Luzzi R., Vasconcellos A.R. and Ramos J.G., Statistical Foundations of Irreversible
Thermodynamics, Teubner, Leipzig, 2001
Luzzi R., Vasconcellos A.R. and Ramos J.G., Predictive Statistical Mechanics: A
Non-equilibrium Ensemble Formalism, Kluwer, Dordrecht, 2002
Lyapounov M., Stability of Motion, Academic, New York, 1966 (English translation)
Ma S.K., Statistical Mechanics, World Scientific, Singapore, 1985
Madruga S., P´erez-Garc´ıa C. and Lebon G., Convective Instabilities in two laterally
superposed horizontal layers heated, Phys. Rev. E 68 (2003) 041607, 1–12
Magnasco M.O., Molecular combustion motors, Phys. Rev. Lett. 72 (1994) 2656–2659
Mahan G., Sales B. and Sharp J., Thermoelectric materials: new approaches to an
old problem, Phys. Today 50 (1997) 42–47
Mandel J., Propri´et´es M´ecaniques des Mat´eriaux, Eyrolles, Paris, 1978
Maugin G., The Thermodynamics of Nonlinear Irreversible Behaviors, World Scien-
tific, Singapore, 1999
Maugin G. and Muschik W., Thermodynamics with internal variables, J. Non-Equilib.
Thermodyn. 19 (1994) 217–289
Maurer M.J. and Thomson H.A., Non-Fourier effects at high heat flux, J. Heat Trans-
fer 95 (1973) 284–286
Maxwell J.C., On the dynamical theory of gases, Philos. Trans. R. Soc. Lond. 157
(1867) 49–88
McDougall J.T. and Turner J.S., Influence of cross-diffusion on ‘finger’ double-
diffusive convection, Nature 299 (1982) 812–814
McQuarrie D.A., Statistical Mechanics, Harper and Row, New York, 1976
Meixner J., Zur thermodynamik der irreversiblen prozesse in gasen mit chemisch
reagierenden, dissozierenden und anregbaren komponenten, Ann. Phys. (Leipzig)
43 (1943) 244–269
Meixner J., The entropy problem in thermodynamics of processes, Rheol. Acta 12
(1973a) 272–283
Meixner J., On the Foundations of Thermodynamics of Processes, in A Critical Re-
view of Thermodynamics (Stuart E., Gal-Or B. and Brainard A., eds.), Mono
Book Corp., Baltimore, 1973b
Meixner J. and Reik H.G., Thermodynamik der Irreversible Prozesse, in Handbuch
der Physik, Bd 3/ II, Springer, Berlin Heidelberg New York, 1959
Mongiov`ı M.S., Extended irreversible thermodynamics of liquid helium II, Phys. Rev.
B 48 (1993), 6276–6289
Mongiov`ı M. S., Proposed measurements of the small entropy carried by the superfluid
component of liquid helium II, Phys. Rev. B. 63 (2001) 12501–12510
Morse P.M. and Feshbach H., Mathematical Methods of Theoretical Physics, McGraw
Hill, New York, 1953
Movchan A., The direct method of Lyapounov in stability problems of elastic systems,
J. Appl. Math. Mech. 23 (1959) 686–700
M¨ uller I., On the frame dependence of stress and heat flux, Arch. Rat. Mech. Anal.
45 (1972) 241
M¨ uller I. and Ruggeri T., Rational Extended Thermodynamics, 2nd ed., Springer,
Berlin Heidelberg New York, 1998
Muschik W., Empirical foundation and axiomatic treatment of non-equilibrium tem-
perature, Arch. Rat. Mech. Anal. 66 (1977) 379–401
314 References
Muschik W., Aspects of Non-Equilibrium Thermodynamics, World Scientific,
Singapore, 1990
Nelson P., Biological Physics. Energy, Information, Life, Freeman, New York, 2004
Nicolis G. and Nicolis C., On the entropy balance of the earth-atmosphere system,
Q. J. R. Meteorol. Soc. 106 (1980) 691–706
Nicolis G. and Prigogine I., Self-Organization in Nonequilibrium Systems, Wiley, New
York, 1977
Nicolis G. and Prigogine I., Exploring Complexity, W.H. Freeman, New York, 1989
Nield D.A., Surface tension and buoyancy effects in cellular convection, J. Fluid Mech.
19 (1964) 341–353
Noll W., The Foundations of Mechanics and Thermodynamics, Springer, Berlin, 1974
Nettleton R.E. and Sobolev S.L., Applications of extended thermodynamics to
chemical, rheological and transport processes: a special survey, J. Non-Equilib.
Thermodyn. 20 (1995) 200–229, 297–331
Onsager L., Reciprocal relations in irreversible processes, Phys. Rev. 37 (1931) 405–
426; 38 (1931) 2265–2279
Onuki A., Phase transitions of fluids in shear flow, J. Phys.: Condens. Mat. 9 (1997)
6119–6157
Onuki A., Phase Transitions Dynamics, Cambridge University Press, Cambridge,
2002
¨
Ottinger H.C., A thermodynamically admissible reptation model for fast flows of
entangled polymers, J. Rheol. 43 (1999a) 1461–1493
¨
Ottinger H.C., Thermodynamically admissible equations for causal dissipative cos-
mology, galaxy formation, and transport processes in gravitational collapse, Phys.
Rev. D 60 (1999b) 103507, 1–9
¨
Ottinger H.C., Beyond Equilibrium Thermodynamics, Wiley, Hoboken, 2005
¨
Ottinger H.C., Non-equilibrium thermodynamics of open systems, Phys. Rev. E 73
(2006) 036126
¨
Ottinger H.C. and Grmela M., Dynamics and thermodynamics of complex fluids. II.
Illustrations of a general formalism, Phys. Rev. E 56 (1997) 6633–6655
Ozawa H., Ohmura A., Lorenz R.D. and Pujol T., The second law of thermodynam-
ics and the global climate system: a review of the maximum entropy production
system, Rev. Geophys. 41 (2003) 4/1018
Palm E., Nonlinear thermal convection, Annu. Rev. Fluid Mech. 7 (1975) 39–61
Paltridge G.W., Thermodynamic dissipation and the global climate system, Q. J. R.
Meteorol. Soc. 107 (1981) 531–547
Parmentier P., Regnier V. and Lebon G., Buoyant-thermocapillary instabilities in
medium Pr number fluid layers subject to a horizontal temperature gradient, Int.
J. Heat Mass Transfer 36 (1993) 2417–2427
Pasmanter R.A., Bedeaux D. and Mazur P., The ballast resistor: an electro-thermal
instability in a conducting wire II; fluctuations around homogeneous stationary
states, Physica A 90 (1978) 151–163
Pearson J.R.A., On convection cells induced by surface tension, J. Fluid Mech. 4
(1958) 489–500
Peixoto J.P. and Oort A.H., The physics of climate, Rev. Mod. Phys. 56 (1984)
365–429
Ponter A.S., Bataille J. and Kestin J., A thermodynamic model for the time inde-
pendent plastic deformation of solids, J. M´ecanique 18 (1978) 511–539
Prigogine I., Etude thermodynamique des ph´enom`enes irr´eversibles, Desoer, Li`ege,
1947
Prigogine I., Introduction to Thermodynamics of Irreversible Processes, Interscience,
New York, 1961
Prigogine I., La fin des certitudes, O. Jacob, Paris, 1996
References 315
Prigogine I. and Lefever R., Symmetry breaking instabilities in dissipative systems,
J. Chem. Phys. 48 (1968) 1695–1700
Prigogine I. and Stengers I., La nouvelle alliance, Gallimard, Paris, 1979
Pritchard A.J., A study of the classical problem of hydrodynamic stability, J. Inst.
Math. Applics. 4 (1968) 78–93
Pujol T. and Llebot J.E., Extremal principle of entropy production in the climate
system, Q. J. R. Meteorol. Soc. 125 (1999) 79–90
Rayleigh Lord, Scientific Papers, vol. VI, Cambridge University Press, Cambridge,
1920
Rebhan E., Efficiency of non-ideal Carnot engines with friction and heat losses, Am.
J. Phys. 70 (2002) 1143–1149
Regnier V., Dauby P.C. and Lebon G., Linear and nonlinear Rayleigh–B´enard–
Marangoni instability with surface deformations, Phys. Fluids 12 (2000) 2787–
2799
Reguera D., Mesoscopic nonequilibrium kinetics of nucleation processes, J. Non-
Equilib. Thermodyn. 29 (2004) 327–344
Reguera D., Rub´ı J.M. and Vilar J.M., The mesoscopic dynamics of thermodynamic
systems, J. Phys. Chem. B 109 (2005) 21502–21515
Reichl L.E., A Modern Course in Statistical Physics, University Texas Press, Austin,
1998
Rivlin R., Forty Years of Non-Linear Continuum Mechanics, in Proceedings of the 9th
International Congress on Rheology, Mexico, 1984
Rosenblat S., Davis S.H. and Homsy G.M., Nonlinear Marangoni convection in
bounded layers, J. Fluid Mech. 120 (1982) 91–138
Rub´ı J.M., The non-equilibrium thermodynamics approach to the dynamics of meso-
scopic systems, J. Non-Equilib. Thermodyn. 29 (2004) 315–325
Rub´ı J.M. and Casas-V´azquez J., Thermodynamical aspects of micropolar fluids. A
non-linear approach, J. Non-Equilib. Thermodyn. 5 (1980) 155–164
Ruelle D., Chance and Chaos, Princeton University Press, Princeton, 1991
Salamon P. and Sieniutycz S. (eds.), Finite-Time Thermodynamics and Optimization,
Taylor and Francis, New York, 1991
Salamon P., Nitzan A., Andresen B. and Berry R.S., Minimum entropy production
and the optimization of heat engines, Phys. Rev. A 21 (1980) 2115–2129
Scriven L.E. and Sternling C.V., On cellular convection driven by surface tension
gradients: effect of mean surface tension and viscosity, J. Fluid Mech. 19 (1964)
321–340
Segel L., Non-linear hydrodynamic stability theory and its application to thermal
convection and curved flows, in Non-Equilibrium Thermodynamics, Variational
Techniques and Stability (Donelly R.F., Herman R. and Prigogine I., eds.), pp
165–197, University of Chicago Press, Chicago, 1966
Serrin J., Conceptual analysis of the classical second law of thermodynamics, Arch.
Rat. Mech. Anal. 70 (1979) 355–371
Shannon C., A mathematical theory of communication, Bell System Tech. J. 27
(1948) 379–423, 623–656
Shl¨ uter A., Lortz D. and Busse F.H., On the stability of steady finite amplitude
convection, J. Fluid Mech. 23 (1965) 129–144
Sidoroff F.J., Variables internes en visco´elasticit´e, J. M´ecanique 14 (1975) 545–566,
571–595; 15 (1976) 95–118
Sieniutycz S., Conservation Laws in Variational Thermohydrodynamics, Kluwer,
Dordrecht, 1994
Sieniutycz S. and Farkas H. (eds.), Variational and Extremum Principles in Macro-
scopic Systems, Elsevier Science, Oxford, 2004
Sieniutycz S. and Salamon P. (eds.), Extended Thermodynamic Systems, Taylor and
Francis, New York, 1992
316 References
Silhavy M., The Mechanics and Thermodynamics of Continuous Media, Springer,
Berlin Heidelberg New York, 1997
Snider R.F. and Lewchuk K.S., Irreversible thermodynamics of a fluid system with
spin, J. Chem. Phys. 46 (1967) 3163–3172
Snyder G.J. and Ursell T.S., Thermoelectric efficiency and compatibility, Phys. Rev.
Lett. 91 (2003) 148301
Sparrow C., The Lorenz Equations: Bifurcations, Chaos and Strange Attractors,
Springer, Berlin Heidelberg New York, 1982
Spencer A.J.M. and Rivlin R.S., Further results in the theory of matrix polynomials,
Arch. Rat. Mech. Anal. 4 (1959) 214–230
Stuart J.T., On the non-linear mechanics of hydrodynamic instability, J. Fluid Mech.
4 (1958) 1–21
Swift J. and Hohenberg P.C., Hydrodynamic fluctuation at the convective instability,
Phys. Rev. A 15 (1977) 319–328
Thom R., Stabilit´e structurelle et morphog´en`ese, Benjam´ın, Reading, MA, 1972
Tisza L. and Manning I., Fluctuations and irreversible thermodynamics, Phys. Rev.
105 (1957) 1695–1704
Truesdell C., Thermodynamics for Beginners, in Irreversible Aspects of Continuum
Mechanics (Parkus H. and Sedov L., eds.), Springer, Berlin Heidelberg New York,
1968
Truesdell C., Rational Thermodynamics, 2nd ed., Springer, Berlin Heidelberg New
York, 1984
Truesdell C. and Noll W., The Non-Linear Field Theories, in Handbuch der Physik,
Bd. III/3, Springer, Berlin Heidelberg New York, 1965
Truesdell C. and Toupin R., The Classical Field Theories, in Handbuch der Physik,
Bd. III/1, Springer, Berlin Heidelberg New York, 1960
Tsallis C., Some thoughts on theoretical physics, Physica A 344 (2004) 718–736
Turcotte D.L., Fractals and Chaos in Geology and Geophysics, Cambridge University
Press, Cambridge, 1992
Turing A., The chemical basis of morphogenesis, Philos. Trans. R. Soc. Lond. B 237
(1952) 37–72
Tzou D.Y., Macro-to-Microscale Heat Transfer. The Lagging Behaviour, Taylor and
Francis, New York, 1997
Valanis K.C., A gradient theory of internal variables, Acta Mechanica 116 (1996)
1–14
Valenti A., Torrisi M. and Lebon G., Shock waves in crystalline dielectrics at low
temperature, J. Phys. Condens. Matter 14 (2002) 3553–3564
Van den Broeck C., Taylor dispersion revisited, Physica A 168 (1990) 677–696
Van den Broeck C., Thermodynamic efficiency at maximum power, Phys. Rev. Lett.
95 (2005) 190602
Van Kampen N.G., Views of a Physicist, World Scientific, Singapore, 2000
Vernotte P., La veritable equation de la chaleur, Compt. Rend. Acad. Sci. Paris 247
(1958) 2103–2107
Veronis G., Cellular convection with a finite amplitude in a rotating fluid, J. Fluid
Mech. 5 (1959) 401–435
Vidal C., Dewel G. and Borckmans P., Au del` a de l’´equilibre, Hermann, Paris, 1994
Vilar J.M.G. and Rub´ı J.M., Thermodynamics beyond local equilibrium, Proc. Natl
Acad. Sci. USA 98 (2001) 11081–11084
de Vos A., Endoreversible Thermodynamics of Solar Energy Conversion, Oxford Uni-
versity Press, Oxford, 1992
Westerhoff N.V. and van Dam K., Thermodynamics and Control of Biological Free-
Energy Transduction, Elsevier, Amsterdam, 1987
Wilmanski K., Thermodynamics of Continua, Springer, Berlin Heidelberg New York,
1998
References 317
Woods L., The Thermodynamics of Fluid Systems, Clarendon, Oxford, 1975
Woods L., The bogus axioms of continuum mechanics, Bull. Inst. Math. Appl. 17
(1981) 98–102; 18 (1982) 64–67
Zemansky M.W., Heat and Thermodynamics, 5th ed., McGraw-Hill, New York, 1968
Further Readings
There exists a multiplicity of textbooks and popular books on the subject and it would
be unrealistic to mention all of them. The list given below is therefore limited and,
of course, subjective.
Atkins P.W., The Second Law, Scientific American Library, W.H. Freeman, New York,
1984
Bailyn M., A Survey of Thermodynamics, AIP, New York, 1993
Coveney P. and Highfield R., The Arrow of Time, Flamingo, London, 1990
Davies P.C.W., The Physics of Time Asymmetry, University of California Press,
Berkeley, 1974
Denbigh K. and Denbigh J., Entropy in Relation with Incomplete Knowledge,
Cambridge University Press, Cambridge, 1985
Feynman R.P. and Weinberg S., Elementary Particles and the Laws of Physics,
Cambridge University Press, Cambridge, 1987
Feynman R.P., Leighton R.B. and Sands M., The Feynman Lectures on Physics,
Addison-Wesley, 1963 (Chaps. 39–46)
Goodwin B., How the Leopard Changed Its Spots: The Evolution of Complexity,
Weidenfeld and Nicolson, London, 1994
Harman P.M., Energy, Force and Matter: The Conceptual Development of
Nineteenth-Century Physics, Cambridge University Press, Cambridge, 1982
Kestin J., A Course in Thermodynamics, 2 vols., Hemisphere, Washington, 1979
Longair M.S., Theoretical Concepts in Physics: An Alternative Views of Theoretical
Reasoning in Physics, 2nd ed., Cambridge University Press, Cambridge, 2003
Maxwell J.C., Theory of Heat, Dover, New York, 2001
Pippard A.B., The Elements of Classical Thermodynamics, Cambridge University
Press, London, 1957
Prigogine I., From Being to Becoming, Freeman, San Francisco, 1980
Prigogine I. and Stengers I., Entre le temps et l’´eternit´e, Fayard, Paris, 1988
Segr`e G., Einstein’s Refrigerator, Penguin Books, London, 2004
Tisza L. and Manning T., Fluctuations and irreversible thermodynamics, Phys. Rev.
105 (1957) 1695–1704
Vidal Ch. and Lemarchand H., La r´eaction cr´eatrice, Herman, Paris, 1988
Walgraef D., Spatio-Temporal Pattern Formation, Springer, Berlin Heidelberg New
York, 1997
Winfree A., The Geometry of Biological Time, Springer, Berlin Heidelberg New York,
1980
Winfree A., The Timing of Biological Clocks, Scientific American Library, Freeman,
New York, 1987a
Winfree A., When Time Breaks Down, Princeton University Press, Princeton, 1987b
319
Index
absolute temperature, 2, 9, 10, 14, 40,
233, 250
accompanying state, 216, 217, 220, 222,
225, 233
active transport, 83, 91, 103, 106, 111
adiabatic, 217
adiabatic wall, 5, 6, 18
affinity, 30, 32, 33, 47, 93, 94, 104, 106,
110, 219, 222, 297
ageing, 44
albedo, 125–128, 133
amplitude method, 136, 139, 156
angular velocity, 49, 67, 138, 158, 159,
161, 175, 242, 243, 251, 276, 285
anisotropic systems, 39, 65, 138
arrow of time, 43
asymptotically stable, 135, 164
atmosphere, 82, 125, 126, 128, 132, 144
autocatalytic reaction, 108, 111, 163,
169, 177
availability, 124, 280
balance equations, 45, 53, 59, 60, 80, 94,
143, 170, 203, 216, 219, 225, 238,
239, 243, 244, 247, 250, 253, 261,
275, 276
ballast resistor, 171, 173
ballistic regime, 182, 191, 192, 195
barycentric velocity, 45, 59, 79, 224, 278
B´enard–Marangoni’s instability, 145,
158, 176
Belousov–Zhabotinsky reaction, 166
bifurcation, 135, 141, 157, 162, 171
biology, 38, 44, 54, 91, 100, 107, 109,
162, 169, 305
black body, 132
Boltzmann–Planck’s formula, 280
boundary conditions, 51–54, 57, 58, 144,
148, 149, 151, 152, 155, 159, 160,
175, 176, 208, 219, 250, 254, 268
Boussinesq approximation, 147, 158
Brownian motion, 106, 107, 235, 285,
293, 298, 300
Brownian Motors, 106
Brusselator models, 163
bulk viscosity, 62, 199, 248, 258
buoyancy–surface-tension, 177
Carnot cycle, 10, 34, 114
catalyses, 107
Cattaneo equation, 191–193, 211, 240
causality, 58
chaos, 44, 136, 153, 303
chemical instabilities, 162, 172
chemical kinetics, 91, 99, 166
chemical potential, 4, 14–16, 19, 28, 30,
40, 59, 70, 80, 83, 84, 93, 202, 203,
213, 225, 231, 268, 278, 290, 291,
293, 297
chemical reactions, 5, 14, 23, 29, 31, 38,
39, 46, 54, 63, 65, 79, 83, 91, 96
classical irreversible thermodynamics,
37, 44, 58, 63, 65, 72, 99, 179,
198, 203, 216, 220, 233, 241, 276,
292–294, 304
Clausius’ inequality, 12, 292
Clausius–Duhem’s inequality, 220, 234,
238, 239, 241, 243, 244, 246, 248,
249, 251, 253, 257–259
climatic changes, 126
clouds, 125, 127
coefficient of thermal expansion, 17, 145,
282
colloidal suspensions, 227, 234
321
322 Index
complex fluids, 224, 226, 227, 275
configuration tensor, 200
conservation laws, 51
constitutive relations, 63, 99, 219, 225,
233, 238, 240–243, 247–249, 251,
253, 272, 304
constrained equilibrium, 218, 221
continued-fraction, 195, 196, 210
control parameter, 135, 138, 139, 163,
174
convected time derivatives, 261
corotational derivative, 261
Couette flow, 158, 162, 177
coupled chemical reactions, 96, 100
coupled processes, 69, 91, 101, 103
critical threshold, 136, 138, 139
critical wave number, 151, 156, 161
Curie’s law, 47, 48
Curzon–Ahlborn’s engine, 129, 130
cyclical chemical reactions, 97
Deborah number, 40, 180, 218
degeneracy conditions, 266, 278
degree of advancement, 29, 92, 95, 96,
234
degree of coupling, 101
dendrites, 172
detailed balance, 97–99, 112
diffusion, 18, 32, 38, 51, 54, 57–59, 63
diffusion coefficient, 38, 49, 60, 67, 81,
89, 168, 276, 294, 297, 300
diffusion flux, 38, 45, 59, 67, 89, 108,
228, 274, 293, 294, 296
dislocations, 216, 223
disorder, 3, 11, 174
dispersion relation, 150, 193, 206
dissipation potential, 265, 266, 269, 270,
272, 274–276
dissipative bracket, 269, 277
dissipative structures, 136
Dufour’s effect, 82
dumbbells, 224
Earth’s Energy Balance, 125
ecology, 53
economy, 91, 205
efficiency, 11, 43, 70, 72, 76–78, 88, 91,
92, 100–103
Einstein’s relation, 67
Einstein’s summation convention, 46,
244
Einstein’s theory of fluctuations, 279
electric conductivity, 38
electric current, 38, 47, 73, 75–77, 87,
88, 171, 181, 203, 213, 285
emissivity, 66, 125, 127, 132
empirical temperature, 5, 9, 10
endoreversible heat engine, 117
enthalpy, 7, 8, 21, 23, 24, 27, 31, 80, 232
entropy, 2–4, 8, 10–12, 15, 18, 20, 21, 23
entropy balance, 41, 42, 46, 94, 128, 129,
231
entropy flux, 42, 61, 66, 71, 72, 80, 87,
94, 130, 190, 198, 209, 220, 223,
225, 230, 231, 234, 239, 252, 256,
257, 277
entropy production, 12, 42, 43, 46, 49,
51, 52, 55, 56, 58, 61–63, 65
enzymes, 103
equation of state, 6, 15, 31, 35, 55, 148,
222
equilibrium, 1, 3–9, 11
equilibrium constant, 92, 93
(local) equilibrium, 191, 257, 291
equipresence, 241, 243, 247, 252
Euler’s relation, 16, 21
Eulerian time derivative, 52
evolution equations, 44, 46, 61–63, 70,
79, 136, 180, 189, 192, 194, 197–199
exchange of stability, 138
exothermic reactions, 92
extended irreversible thermodynamics,
58, 179, 188, 189, 200, 202, 207,
215, 232, 233, 240, 252, 276, 278,
286, 304
extensive variables, 4, 15, 19, 217, 220,
288, 290, 291
extremum principles, 21, 51
Fick’s law, 38, 60, 81, 231, 272, 295
figure of merit, 70, 77, 88
finite-time thermodynamics, 113, 125,
128
first law, 2, 3, 5–8, 11, 14, 19, 22, 31, 92,
114, 117, 187, 217
fluctuation–dissipation theorem, 279,
285, 286
fluctuations, 4, 24–26, 40, 49, 50, 64,
105, 135, 153, 174, 206, 228, 279,
280
fluid flows, 5, 54, 66, 162, 200, 261
fluxes, 4, 38, 39, 45–48, 50, 61, 63, 64,
68, 69, 77, 80, 85, 88, 101, 107
Fokker–Planck’s equation, 236, 295, 297,
298
Fourier transforms, 194
Index 323
Fourier’s law, 38, 56, 65, 66, 68, 76,
181–183, 185, 187, 190, 193, 195,
211, 254, 257
Fourier’s law with, 194
frame-indifference, 201, 226, 242, 247,
251, 276
friction losses, 120
fundamental relations, 15, 19, 21
generalized hydrodynamics, 64, 212
GENERIC formalism, 266, 274, 277
Gibbs’ equation, 1, 14–16, 32, 40, 45, 59,
62, 63, 70, 72, 80, 189, 191, 197,
203, 213, 219, 222, 225, 230, 234,
235, 238, 244, 248, 249, 256, 281,
290, 293, 298
Gibbs’ free energy, 21, 23, 24, 27, 30, 51,
213
Gibbs–Duhem’s relation, 16, 17, 60, 80
Ginzburg–Landau’s equation, 153
global warming, 125–127, 132
Goldmann equation, 90
Green–Kubo’s formula, 287
greenhouse effect, 126, 127, 132
Guyer–Krumhansl equation, 193, 209,
211
gyroscopic forces, 261, 276
Hall effect, 71
Hamilton equations, 262, 266
Hamiltonian formalisms, 200
harmonic oscillator, 263
heat, 2, 3, 5–9, 11, 14, 18, 19, 21, 26
heat capacity at constant pressure, 17,
282
heat capacity at constant volume, 17, 25
heat conduction, 38, 50, 51, 54, 76, 77,
181, 195, 211, 232, 234, 240, 251,
253, 254, 258, 259, 298, 304
heat conductivity, 49, 55, 56, 58, 62, 63,
65, 72, 77, 81, 149, 170, 176, 181,
191, 194, 196, 232, 245, 248, 256,
257, 287
heat diffusivity, 56, 66, 183
heat engines, 10, 88, 101, 103, 113, 114,
123, 131
heat flux, 35, 37, 38, 45, 47, 50, 54, 55,
64, 73, 76, 80, 93, 170, 181
heat of reaction, 31–33
Helmholtz’s free energy, 23, 24, 27, 36,
267, 270, 276, 281
hexagonal patterns, 156
history, 237, 238, 240, 241
Hopf bifurcation, 138
hydrodynamics, 38, 60, 62, 63, 136, 162,
167, 173, 208, 210, 247, 248, 251,
266–268, 270, 275, 277
hyperbolic equations, 184
hysteresis, 143
ideal gas, 13, 30, 31, 34, 35, 93, 114, 117,
195, 287, 299
incompressible fluids, 62
inertial effects, 116, 293, 298
information, 3, 15, 43, 64, 96, 109, 113,
122, 128, 151, 156, 203, 205, 264,
272, 276, 281, 304
instabilities, 44, 135, 136, 145, 158, 162,
170, 173, 212, 279
intensive variables, 4, 15, 16, 19, 24, 41,
46, 52, 65, 80, 290, 292
internal energy, 2, 3, 6, 7, 12, 20, 22, 35,
38–40, 45, 54, 60, 70, 79, 92
internal variables, 179, 200, 215,
217–219, 221, 223
irreversibility, 43, 117, 120, 135
irreversible processes, 5, 12, 15, 24–26,
34, 37, 38, 42–44, 46, 50, 61, 63,
79, 95, 115, 215, 251, 265, 274, 277,
298
isothermal compressibility, 17, 25, 41,
52, 282, 299
isotope separation, 82
isotropic systems, 47, 197
isotropic tensors, 47, 48, 248, 258
Jacobi’s identity, 264, 265, 276, 278
Joule dissipation, 77
Kelvin relation, 73, 76
Kelvin–Planck formulation of the second
law, 114
Kelvin–Voigt model, 223
kinetic constants, 97, 109–111
Knudsen number, 180, 192, 195
Lagrange multipliers, 253, 255
Lagrangian time derivative, 41
Landau’s equation, 141
lasers, 70
Le Chatelier’s principle, 26, 28, 32, 36
Legendre transformations, 19, 24, 27
light scattering, 199, 208, 292
limit cycle, 164, 166
Liu’s Lagrange Multipliers, 253
local equilibrium, 39–41, 45, 63, 179,
185, 186, 189, 191, 197, 202, 207,
324 Index
215–217, 230, 252, 257, 288, 293,
296, 298, 301
Lorenz model, 153, 177
Lotka–Volterra model, 163, 166, 177
Lyapounov function, 53, 58, 270, 290
Lyapounov’s functional, 139
magnetic fields, 39, 219
Marangoni effect, 144, 177
Marangoni number, 155, 156, 176, 177
marginal stability, 136, 138, 147, 150,
156, 161
mass action law, 110
mass fraction, 40, 45, 59, 60, 70, 80, 81,
92, 96, 228, 271, 273, 278
mass transport, 37, 54, 65, 79, 100, 103,
104, 169, 277, 278
Massieu–Planck functions, 21
materials sciences, 69, 78
Maxwell’s relations, 17, 35
Maxwell–Cattaneo’s equations, 199
mean free path, 70, 192
mechanical filtration coefficient, 85
membranes, 54, 69, 70, 83, 85–87, 89,
102, 292, 297
memory, 179, 181, 206, 207, 219, 237,
240, 241, 247, 251, 257, 286, 287,
300, 304
mesoscopic, 275, 277, 292
metastable, 142
Michaelis–Menten’s, 109, 110
microelectronic devices, 38, 181, 202,
207
micropolar fluids, 67
microscopic reversibility, 49, 64, 284
minimum entropy production, 51, 53,
58, 68, 124, 131, 134
mixtures, 27, 63, 70, 79, 81, 163, 170,
228, 231, 270, 273, 278
mole fraction, 31, 93
molecular motors, 91, 97, 100, 103, 105,
106, 110
morphogenesis, 108, 168
multi-component systems, 28
nano-systems, 195, 205
Navier–Stokes’ equation, 62, 63, 160,
270
Nernst equation, 89
Neumann–Duhamel’s relation, 245
Newton’s cooling law, 65, 149, 176
Newton–Stokes’ law, 37, 199
non-equilibrium entropy, 180, 207, 257,
279, 289, 304
non-equilibrium temperature, 189, 190,
197, 250, 255, 257, 277, 291, 295,
301
Non-Fickian diffusion, 273, 274, 278
non-local effects, 179, 192, 193, 205, 207,
219, 233
normal modes, 136, 137
normal stresses, 201
Ohm’s law, 38, 69, 71, 203
osmotic pressure, 83–85
Other sources, 128
overstability, 138
oxidative phosphorylation, 100, 103
parabolic differential equation, 60, 182
pattern formation, 136, 169, 172, 174
Peltier’s effect, 73, 74, 76, 87
permeability coefficient, 85, 87
phase speed, 183
phase transitions, 135, 143, 215
phenomenological coefficients, 47, 49,
52, 53, 62, 64, 65, 68, 72, 81, 85,
87, 101, 102, 104, 111, 226, 231,
251, 294, 298
phenomenological equations, 48, 62, 72,
98, 100
phonons, 70, 182, 192–195, 210
photovoltaic cells, 114, 132
Piola–Kirchhoff stress tensor, 246
Poisson brackets, 266, 275
polymers, 40, 63, 170, 180, 182, 205,
207, 208, 232, 274, 298
power, 2, 70, 73, 74, 76, 78, 88, 110, 113,
115, 117–121
power lasers, 69
Poynting–Thomson, 223, 234
Prandtl number, 147, 149
pressure tensor, 45, 61, 64, 67, 181, 193,
200, 201, 205, 206, 212, 213, 238,
270, 276
prey–predator system, 163
radiation, 35, 66, 125, 127, 128, 132, 238
rational extended thermodynamics, 200,
253, 254
rational thermodynamics, 179, 200, 219,
233, 237, 238, 249, 251, 253, 254,
259, 275–277, 304
Rayleigh number, 138, 147, 150, 151,
154, 155, 161, 174, 175, 177
Rayleigh–B´enard’s instability, 145, 147,
150, 152, 161, 174, 175
reaction–diffusion, 136, 208
Index 325
reciprocal relations, 49, 50, 56, 64, 69,
72, 87, 91, 96, 97, 99, 104, 231,
277, 279, 282, 285
reflection coefficient, 86
refrigerators, 75, 114, 127, 131
relaxation time, 40, 95, 146, 170, 179,
180, 182, 184, 186, 188, 191, 192,
194, 195, 199, 200
reversible processes, 1, 5, 7, 9, 12, 34,
41–43, 46, 113, 115, 303
Reynolds number, 138
rheology, 201, 215, 223, 227, 251, 252,
261, 277, 305
roll pattern, 145
salt fingers, 158, 169, 171
second law, 2, 3, 5, 8, 11, 12, 14, 18, 22,
23, 36, 39, 42
second sound, 183, 184, 192, 194, 208,
212
Seebeck’s effect, 73, 74
self-adjoint problems, 138
self-organization, 91, 136, 163, 169, 174
semiconductors, 69, 74, 75, 79
shortwave, 125
silicon diode, 204
solar energy, 114, 129
Soret’s effect, 82
sources, 79, 115, 117, 126, 127, 223, 238
stability of equilibrium, 1, 24, 26, 28, 30,
52, 60, 200, 270
stable steady states, 290
state, 1, 3–7, 10, 11, 13, 15, 16, 18
state variables, 3, 4, 6, 15, 24, 25, 40, 45,
49–51, 63, 64, 79, 180
steady states, 181, 202, 211, 288, 300
Stefan–Boltzmann law, 35, 131
stochastic noise, 289
stoichiometric coefficients, 29, 97
Stokes’ law, 270
strain tensor, 217, 240, 243, 245, 246
stress tensor, 61, 155, 176, 223, 225, 226,
229, 231, 235, 238, 240, 242, 244,
246, 248, 252, 258, 275
subcritical instability, 142, 143, 157
supercritical stability, 140
superfluids, 63, 180, 182, 205, 208, 234
surface tension, 140, 142, 144, 154–156,
176
surface-tension, 158
suspensions, 180, 208, 227, 228, 231–234
symplectic manifold, 264
Taylor number, 138, 161, 175
Taylor’s instability, 158, 161, 162
telegrapher’s equation, 183, 184
temperature equation, 56, 259
thermal convection, 136, 143, 144, 158
thermal diffusion, 81, 82, 89, 146, 147,
228
thermal waves, 184, 191–193, 210, 211
thermocouples, 73, 74
thermodiffusion, 18, 38, 54, 69, 70
thermodynamic degrees of freedom, 16
thermodynamic forces, 46, 47, 52, 54,
64, 68, 69, 111, 179, 283, 287
thermodynamic potentials, 19, 21, 24,
27, 35, 135, 233, 249, 261
thermoelasticity, 243
thermoelectric conversion, 76
thermoelectric generators, 72, 76, 102
thermoelectricity, 38, 47, 54, 69, 73
Thermophoresis, 89
Third Law, 14
Thomson effects, 38
time reversal, 43, 49, 226, 264, 269–271,
285
time-reversal, 284, 294
transcritical bifurcation, 142
transport equations, 37, 47, 61, 179, 180,
192, 205
triangular chemical, 91
Turing Structures, 167
Turing’s instability, 173
two temperatures, 73
ultrafiltration coefficient, 85
ultrasound propagation, 40, 64, 180, 199
upper-convected Maxwell model, 201
van’t Hoff Relation, 31, 32
(extensive) variables, 13
variational principles, 51, 53, 68
velocity of reaction, 29, 92, 94, 110
viscoelasticity, 221
viscosity, 62, 63, 67, 68, 79, 105, 117,
140, 142, 146, 147, 158, 175, 199,
201, 212, 225, 234, 248, 251, 269,
273, 276, 286
viscous fluids, 59, 181, 270
Volterra derivative, 265
wave equation, 183
Wien’s law of radiation, 125
wind energy, 114, 130
work, 2, 6–8, 10, 14, 21, 24, 34, 36, 37,
46, 54, 57, 92, 100
Zeroth Law, 5

Understanding Non-equilibrium Thermodynamics

G. Lebon • D. Jou • J. Casas-Vázquez

Understanding Non-equilibrium Thermodynamics

Foundations, Applications, Frontiers

123

Prof. Dr. David Jou Universitat Autònoma de Barcelona Dept. Fisica - Edifici Cc Grup Fisica Estadistica Bellaterra 08193 Catalonia, Spain David.Jou@uab.es Prof. Dr. Georgy Lebon Université de Liège Dept. d’Astrophysique G´ ophysique et Oc´ anographie e e B5 Sart Tilman Liege 4000 Belgium g.lebon@ulg.ac.be

Prof. Dr. José Casas-Vázquez Universitat Autònoma de Barcelona Dept. Fisica - Edifici Cc Grup Fisica Estadistica Bellaterra 08193 Catalonia, Spain Jose.Casas@uab.es

Cover image: Thermal radiation leaving the Earth, seen by the EOS-Terra satellite (NASA). Image from www.visibleearth.nasa.gov. Owner: NASA.

ISBN: 978-3-540-74251-7

e-ISBN: 978-3-540-74252-4

Library of Congress Control Number: 2007935452 c 2008 Springer-Verlag Berlin Heidelberg This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable to prosecution under the German Copyright Law. The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Cover design: WMX Design GmbH, Heidelberg Printed on acid-free paper 9 8 7 6 5 4 3 2 1 springer.com

Preface

Our time is characterized by an explosion of information and by an acceleration of knowledge. A book cannot compete with the huge amount of data available on the Web. However, to assimilate all this information, it is necessary to structure our knowledge in a useful conceptual framework. The purpose of the present work is to provide such a structure for students and researchers interested by the current state of the art of non-equilibrium thermodynamics. The main features of the book are a concise and critical presentation of the basic ideas, illustrated by a series of examples, selected not only for their pedagogical value but also for the perspectives offered by recent technological advances. This book is aimed at students and researchers in physics, chemistry, engineering, material sciences, and biology. We have been guided by two apparently antagonistic objectives: generality and simplicity. To make the book accessible to a large audience of nonspecialists, we have decided about a simplified but rigorous presentation. Emphasis is put on the underlying physical background without sacrificing mathematical rigour, the several formalisms being illustrated by a list of examples and problems. All over this work, we have been guided by the formula: “Get the more from the less”, with the purpose to make a maximum of people aware of a maximum of knowledge from a minimum of basic tools. Besides being an introductory text, our objective is to present an overview, as general as possible, of the more recent developments in non-equilibrium thermodynamics, especially beyond the local equilibrium description. This is partially a terra incognita, an unknown land, because basic concepts as temperature, entropy, and the validity of the second law become problematic beyond the local equilibrium hypothesis. The answers provided up to now must be considered as partial and provisional, but are nevertheless worth to be examined. Chapters 1 and 2 are introductory chapters in which the main concepts underlying equilibrium thermodynamics and classical non-equilibrium thermodynamics are stated. The basic notions are discussed with special emphasis on these needed later in this book. V

VI

Preface

Several applications of classical non-equilibrium thermodynamics are presented in Chaps. 3 and 4. These illustrations have not been chosen arbitrarily, but keeping in mind the perspectives opened by recent technological advancements. For instance, advances in material sciences have led to promising possibilities for thermoelectric devices; localized intense laser heating used to make easier the separation of molecules has contributed to a revival of interest in thermodiffusion; chemical reactions are of special interest in biology, in relation with their coupling with active transport across membranes and recent developments of molecular motors. The purpose of Chaps. 5 and 6 is to discuss two particular aspects of classical non-equilibrium thermodynamics which have been the subject of active research during the last decades. Chapter 5 is devoted to finite-time thermodynamics whose main concern is the competition between maximum efficiency and maximum power and its impact on economy and ecology. This classical subject is treated here in an updated form, taking into account the last technological possibilities and challenges, as well as some social concerns. Chapter 6 deals with instabilities and pattern formation; organized structures occur in closed and open systems as a consequence of fluctuations growing far from equilibrium under the action of external forces. Patterns are observed in a multitude of our daily life experiences, like in hydrodynamics, biology, chemistry, electricity, material sciences, or geology. After introducing the mathematical theory of stability, several examples of ordered structures are analysed with a special attention to the celebrated B´nard cells. e Chapters 1–6 may provide a self-consistent basis for a graduate introductory course in non-equilibrium thermodynamics. In the remainder of the book, we go beyond the framework of the classical description and spend some time to address and compare the most recent developments in non-equilibrium thermodynamics. Chapters 7–11 will be of interest for students and researchers, who feel attracted by new scientific projects wherein they may be involved. This second part of the book may provide the basis for an advanced graduate or even postgraduate course on the several trends in contemporary thermodynamics. The coexistence of several schools in non-equilibrium thermodynamics is a reality; it is not a surprise in view of the complexity of most macroscopic systems and the fact that some basic notions as temperature and entropy are not univocally defined outside equilibrium. To appreciate this form of multiculturalism in a positive sense, it is obviously necessary to know what are the foundations of these theories and to which extent they are related. A superficial inspection reveals that some viewpoints are overlapping but none of them is rigorously equivalent to the other. A detailed and complete understanding of the relationship among the diverse schools turns out to be not an easy task. The first difficulty stems from the fact that each approach is associated with a certain insight, we may even say an intuition or feeling that is sometimes rather difficult to apprehend. Also some unavoidable differences in the terminology and the notation do not facilitate the communication. Another

results. 7). Moreover. it is our belief that a good comprehension of these different versions allows for a better and more understandable comprehension of theories whose opportunity was not offered to be discussed here. memory and non-local effects. theories with internal variables (Chap. within the perspectives of improvement and unification. Chapters 7–11 are autonomous and self-consistent. advantages. It could be surprising that the book is completely devoted to macroscopic and mesoscopic aspects and that microscopic theories have been widely omitted. At the end of each chapter is given a list of problems. it is highly recommended to browse through all the chapters to better apprehend the essence and the complementarity of the diverse theories. Comparison of different viewpoints may be helpful for a deeper comprehension and a possible synthesis of the many faces of the theory.Preface VII factor that contributes to the difficulty to reaching a mutual comprehension is that the schools are not frozen in time: they evolve as a consequence of internal dynamics and by contact with others. It is clear that our choice is subjective: we have nevertheless been guided not only by the pedagogical aspect and/or the impact and universality of the different formalisms. 9). One problem was the selection of the main representative ones among the wealth of thermodynamic formalisms. but also to stimulate his interest to solve concrete situations. we hope to convince the reader that macroscopic approaches. Notwithstanding. 7–11 are not only to get rid of the local equilibrium hypothesis. Some of these problems have been . 8). but also by the fact that we had to restrict ourselves. and mesoscopic approaches (Chap. like thermodynamics. it is highly desirable to include as many microscopic results as possible into the macroscopic framework. they have been structured in such a way that they can be read independently of each other and in arbitrary order. Here we have focused our attention on five of them: extended thermodynamics (Chap. rational thermodynamics (Chap.11). Hamiltonian formulation (Chap. Our goal is to contribute to a better understanding among the different schools by discussing their main concepts. The aim is not only to allow the reader to check his understanding. 10). deserve a careful attention and are the seeds of the progress of knowledge. The common points shared by the theories presented in Chaps. However. which is the pillar of the classical theory. Such a comparative study is not found in other textbooks. In each of them. we have tried to save the particular spirit of each theory. Although statistical mechanics appears to be more fashionable than thermodynamics in the eyes of some people and the developments of microscopic methods are challenging. and limitations. we remain convinced that. with the purpose to account for the technological requirements of faster processes and more miniaturized devices. The reasons are that many excellent treatises have been written on microscopic theories and that we decided to keep the volume of the book to a reasonable ratio. but also to propose new phenomenological approaches involving non-linearities.

and of the Direcci´ General de Recerca of the Generalitat of o Catalonia. Jou. and which may be consulted for further investigation. March 2007 e G. J. for the discuse sions on these and related topics for more than 30 years. and in particular M. Li`ge-Bellaterra. Casas-V´zquez a . Lebon. Desaive (Li`ge University). under grants 2001 SGR 00186 and 2005 SGR 00087. Grmela (Montreal University). which are mentioned. Drs. P. We also appreciate our close collaborators for their help and stimulus in research and teaching. We also acknowledge the finantial support of the Direcci´n General de Investigaci´n of the o o Spanish Ministry of Education under grants BFM2003-06003 and FIS200612296-C02-01. We acknowledge many colleagues. Vicen¸ M´ndez and Vicente Ortega-Cejas deserve special gratitude c e for their help in the technical preparation of this book. D.C Dauby and Th. More technical and advanced parts are confined in boxes and can be omitted during a first reading.VIII Preface inspired by recent papers.

. . . . . . . . . .4. . . . . . . . 1. . . . . . . 1. . 1. 1. . . . . . . . . . . .3. . . . 1. . . . . 1. . . . . . . . . . . . . 1 1 3 5 5 6 8 14 14 15 16 16 17 18 19 20 21 24 24 27 27 29 30 31 32 34 34 IX . . . 1. . . . . . . . . .3 Stability of Chemical Equilibrium and Le Chatelier’s Principle . . . . .4 Gibbs’ Equation . . . . . . . . . . . . 1. . . . .6. . 1. . 1. . . . . .4. . . . 1. . . . . . . . . . 1. . . . . . . . . .1 Thermodynamic Potentials . . . .5 The Basic Problem of Equilibrium Thermodynamics . . . . . . . . . . . . . . . .1 Stability of Single Component Systems . . . . . . . . . . . . . . . . . . .6. . . . .4. . .2 Euler’s Relation . . . . . . . . . . . . . .3. . . . . . . . . . . . .4. . . . . .Contents 1 Equilibrium Thermodynamics: A Review . . . . . . . . . . . 1. . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . 1. . . . 1. . . . 1. . .7.2 Heat of Reaction and van’t Hoff Relation . . . . . . . . . . . . . . . . . . . . . . .2 Scope and Definitions . . . . . . . .1 General Equilibrium Conditions . . . . . . . . . . . . 1. . . . 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Gibbs–Duhem’s Relation . . . . . . . . . . . . . . . . . . . . . . . . . 1. . 1. . . . .3. . . . . . . . . . . . . . .4 Some Definitions . . . . . . . .5 Legendre Transformations and Thermodynamic Potentials . . . 1. .1 Fundamental Relations and State Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7. . . . . .6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1. . . 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5. . . . . . . . . . . . . . .7 Equilibrium Chemical Thermodynamics . . . . . . .2 The First Law or Energy Balance . . . . 1. . . 1. . . . . . . .4. . .1 The Early History . . . . . . . . . .1 The Zeroth Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8 Final Comments . . . . . . . . .3 The Fundamental Laws . . . .2 Thermodynamic Potentials and Extremum Principles . .4 The Third Law . . . . . . . . 1.6 Stability of Equilibrium States .3 Stability Criterion of Multi-Component Mixtures . . .5. . . .2 Stability Conditions for the Other Thermodynamic Potentials . . . . . . . . . . . . . . . . . . . . .7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9 Problems . . . . . . . .3 The Second Law . . . . . .

. . . . . .2 Matter Diffusion Under Isothermal and Isobaric Conditions . . . . . . . . 2. . . . . . . . . . . . .1. . . . . . . . . . .6 Problems . . . . . . . and Fluid Flows . . . . . . .1 Curzon–Ahlborn’s Model: Heat Losses . .1 Electrical Conduction . . . . . . . . . . . . . . 3. . . . . . . . . . . . . . . . . . 4. . . . . . . . 5. . . . . . . . . 5.3 Thermodiffusion: Coupling of Heat and Mass Transport . . . . .1. . .5 Autocatalytic Reactions and Diffusion: Morphogenesis . . . . .6. . . . . . . . . . . . . .4. . . . . . . . . . .2. . . . 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 38 39 41 44 50 51 54 54 59 60 63 65 69 70 72 72 76 79 83 83 85 87 91 92 96 96 97 100 102 108 109 113 114 115 120 122 3 4 5 . . . . . . . . . . . . 2. . . . . . . . . . . . . . 3. . . . . 2.6. . . .2 Cyclical Chemical Reactions and Onsager’s Reciprocal Relations . . . . . . . . . . . . . . . . . . . . .1 General Formalism . . .1 Phenomenological Laws . . . . . . . . . . . . . 3. 3. . . . . . . . . .1 Basic Concepts . . . . 4.2 Thermoelectric Effects . . . . . . . . . . . . . .5 Problems . . . . . . . . . .1 Heat Conduction in a Rigid Body . . Chemical Reactions and Molecular Machines . . . . . . . . . . .1 Minimum Entropy Production Principle . . . . . . . . . . . 2. . . . . . . . .2 Economical and Ecological Constraints . . . . . . . . . . . . . . . . . . . .2. . . . . . . .7 Limitations of the Classical Theory of Irreversible Thermodynamics . . . . . . 4. Mass Transport. . . . . . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .X Contents 2 Classical Irreversible Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . . . . . . 3. . . . . . . . . . . . . . . . 2. .6 Applications to Heat Conduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . . 3. . . . . . . . . . . . .3 Efficiency of Energy Transfer . . . . . . . . . . . . . . . . . . . 4. . . . . . . . . . . . .1 Entropy Production . . . . . . . . . . . . . . . . . .2. . . . . . . 2. . . . . .5 Stationary States . . . . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . 3.6. . . . . . . . . .2 Efficiency of Thermoelectric Generators . . .2 Local Equilibrium Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Friction Losses . . . . . Coupled Transport Phenomena . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. . 2. . . . . 4. . .1 One Single Chemical Reaction . . . . . . . . . . .3 Hydrodynamics . . . . . . . .4 General Theory . .4 Chemical Reactions and Mass Transport: Molecular Machines . . . . .3 Entropy Balance . . . . . . . . . . . . . . . . . . . 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. . . .1 The Finite-Time Carnot Cycle . . . . .5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . Finite-Time Thermodynamics . . . .8 Problems . . .2.4 Diffusion Through a Membrane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . . . . 5. . . . . . . .2 Phenomenological Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Coupled Chemical Reactions . . . . . . 3. . . . . . . . . . . . . . . . . 2. .

. . . . . . .2 The Rayleigh–B´nard’s Instability: e A Non-Linear Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7. . . . . . . . . . . . . . .2 Global Warming . 5. . .2 Patterns in Electricity . .1 Earth’s Energy Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 The Rayleigh–B´nard’s Instability: A Linear Theory .3. . . . . . . . . .1 Heat Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6. . . . . . . . . . . . . . . . . . . . . . 6. . . . . 6 Instabilities and Pattern Formation . .5 Final Comments and Perspectives . . .3 Transformation of Solar Heat into Wind Motion . . .3. . . . .1 Temporal Organization in Spatially Homogeneous Systems . 6. . . . . . . . . . . . . . .3. . . . . . . . . . . . 6.5 Chemical Instabilities . . . . .1. . . .2 Spatial Organization in Spatially Heterogeneous Systems . . . . . . . . . . . . . . . . . . . . . . . . 7. . . . . . . . . . . . . . . . . . . . .3 B´nard–Marangoni’s Surface Tension-Driven e Instability . . . . . 6. . . . .3 Rheological Fluids .3 Spatio-Temporal Patterns in Heterogeneous Systems: Turing Structures . . . .2 Non-Linear Approaches . . . . . . . . . . . .4 Taylor’s Instability . . . . . . . . .6 Problems . . 6.1 Fourier’s vs. . . . . . . . . . . . . .5. . . . . . . . . .1. . . . . . . . . .6 Miscellaneous Examples of Pattern Formation . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5. . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7. . . 6. . . . . . . . .1 The Linear Theory of Stability . . .3 Non-Local Terms: From Collision-Dominated Regime to Ballistic Regime . . . . . . .